Generating and editing images from open domain text prompts is a challenging task that heretofore has required expensive and specially trained models. We demonstrate a novel methodology for both tasks which is capable of producing images of high visual quality from text prompts of significant semantic complexity without any training by using a multimodal encoder to guide image generations. We demonstrate on a variety of tasks how using CLIP [37] to guide VQGAN [11] produces higher visual quality outputs than prior, less flexible approaches like DALL·E [38], GLIDE [33] and Open-Edit [24], despite not being trained for the tasks presented. Our code is available in a public repository.

Videos are created to express emotion, exchange information, and share experiences. Video synthesis has intrigued researchers for a long time. Despite the rapid progress driven by advances in visual synthesis, most existing studies focus on improving the frames’ quality and the transitions between them, while little progress has been made in generating longer videos.

In this paper, we present a method that builds on 3D-VQGAN and transformers to generate videos with thousands of frames.

Our evaluation shows that our model trained on 16-frame video clips from standard benchmarks such as UCF101⁠, Sky Time-lapse, and Taichi-HD datasets can generate diverse, coherent, and high-quality long videos. We also showcase conditional extensions of our approach for generating meaningful long videos by incorporating temporal information with text and audio.

Videos and code can be found at https://songweige.github.io/projects/tats/index.html⁠.

The goal of imitation learning is to mimic expert behavior from demonstrations, without access to an explicit reward signal. A popular class of approach infers the (unknown) reward function via inverse reinforcement learning (IRL) followed by maximizing this reward function via reinforcement learning (RL). The policies learned via these approaches are however very brittle in practice and deteriorate quickly even with small test-time perturbations due to compounding errors. We propose Imitation with Planning at Test-time (IMPLANT), a new meta-algorithm for imitation learning that utilizes decision-time planning to correct for compounding errors of any base imitation policy. In contrast to existing approaches, we retain both the imitation policy and the rewards model at decision-time, thereby benefiting from the learning signal of the two components. Empirically, we demonstrate that IMPLANT significantly outperforms benchmark imitation learning approaches on standard control environments and excels at zero-shot generalization when subject to challenging perturbations in test-time dynamics.

We push forward neural network compression research by exploiting a novel challenging task of large-scale conditional generative adversarial networks (GANs) compression. To this end, we propose a gradually shrinking GAN (PPCD-GAN) by introducing progressive pruning residual block (PP-Res) and class-aware distillation.

The PP-Res is an extension of the conventional residual block where each convolutional layer is followed by a learnable mask layer to progressively prune network parameters as training proceeds. The class-aware distillation, on the other hand, enhances the stability of training by transferring immense knowledge from a well-trained teacher model through instructive attention maps.

We train the pruning and distillation processes simultaneously on a well-known GAN architecture in an end-to-end manner. After training, all redundant parameters as well as the mask layers are discarded, yielding a lighter network while retaining the performance.

We comprehensively illustrate, on ImageNet 128×128 dataset, PPCD-GAN reduces up to 5.2× (81%) parameters against state-of-the-arts while keeping better performance.

Pretraining language models with next-token prediction on massive text corpora has delivered phenomenal zero-shot, few-shot, transfer learning and multi-tasking capabilities on both generative and discriminative language tasks. Motivated by this success, we explore a Vector-quantized Image Modeling (VIM) approach that involves pretraining a Transformer to predict rasterized image tokens autoregressively. The discrete image tokens are encoded from a learned Vision-Transformer-based VQGAN (ViT-VQGAN). We first propose multiple improvements over vanilla VQGAN from architecture to codebook learning, yielding better efficiency and reconstruction fidelity. The improved ViT-VQGAN further improves vector-quantized image modeling tasks, including unconditional, class-conditioned image generation and unsupervised representation learning. When trained on ImageNet at 256×256 resolution, we achieve Inception Score (IS) of 175.1 and Fréchet Inception Distance (FID) of 4.17, a dramatic improvement over the vanilla VQGAN, which obtains 70.6 and 17.04 for IS and FID, respectively. Based on ViT-VQGAN and unsupervised pretraining, we further evaluate the pretrained Transformer by averaging intermediate features, similar to Image GPT (iGPT). This ImageNet-pretrained VIM-L significantly beats iGPT-L on linear-probe accuracy from 60.3% to 73.2% for a similar model size. ViM-L also outperforms iGPT-XL which is trained with extra web image data and larger model size.

[Keywords: VQGAN, Vision Transformers, Vector-quantized Image Modeling]

Previous numerical results have shown that a two time-scale update rule (TTUR) using constant learning rates is practically useful for training generative adversarial networks (GANs). Meanwhile, a theoretical analysis of TTUR to find a stationary local Nash equilibrium of a Nash equilibrium problem with two players, a discriminator and a generator, has been given using decaying learning rates. In this paper, we give a theoretical analysis of TTUR using constant learning rates to bridge the gap between theory and practice. In particular, we show that, for TTUR using constant learning rates, the number of steps needed to find a stationary local Nash equilibrium decreases as the batch size increases. We also provide numerical results to support our theoretical analyzes.

Annotating images with pixel-wise labels is a time-consuming and costly process. Recently, DatasetGAN showcased a promising alternative—to synthesize a large labeled dataset via a generative adversarial network (GAN) by exploiting a small set of manually labeled, GAN-generated images. Here, we scale DatasetGAN to ImageNet scale of class diversity.

We take image samples from the class-conditional generative model BigGAN trained on ImageNet, and manually annotate 5 images per class, for all 1k classes. By training an effective feature segmentation architecture on top of BigGAN, we turn BigGAN into a labeled dataset generator. We further show that VQGAN can similarly serve as a dataset generator, leveraging the already annotated data.

We create a new ImageNet benchmark by labeling an additional set of 8k real images and evaluate segmentation performance in a variety of settings. Through an extensive ablation study we show big gains in leveraging a large generated dataset to train different supervised and self-supervised backbone models on pixel-wise tasks. Furthermore, we demonstrate that using our synthesized datasets for pre-training leads to improvements over standard ImageNet pre-training on several downstream datasets, such as PASCAL-VOC, MS-COCO, Cityscapes and chest X-ray, as well as tasks (detection, segmentation). Our benchmark will be made public and maintain a leaderboard for this challenging task.

Project Page: https: /  ​ /  ​nv-tlabs.github.io /  ​big-datasetgan /  ​

The state-of-the-art photo upsampling method, PULSE⁠, demonstrates that a sharp, high-resolution (HR) version of a given low-resolution (LR) input can be obtained by exploring the latent space of generative models. However, mapping an extreme LR input (16²) directly to an HR image (1024²) is too ambiguous to preserve faithful local facial semantics.

In this paper, we propose an enhanced upsampling approach, Pro-PULSE, that addresses the issues of semantic inconsistency and optimization complexity.

Our idea is to learn an encoder that progressively constructs the HR latent codes in the extended 𝑊+ latent space of StyleGAN⁠. This design divides the complex 64× upsampling problem into several steps, and therefore small-scale facial semantics can be inherited from one end to the other.

In particular, we train 2 encoders, the base encoder maps latent vectors in 𝑊 space and serves as a foundation of the HR latent vector, while the second scale-specific encoder performed in 𝑊+ space gradually replaces the previous vector produced by the base encoder at each scale. This process produces intermediate side-outputs, which injects deep supervision into the training of encoder.

Extensive experiments demonstrate superiority over the latest latent space exploration methods, in terms of efficiency, quantitative quality metrics, and qualitative visual results.

Recently, significant progress has been made in learned image and video compression. In particular the usage of Generative Adversarial Networks has lead to impressive results in the low bit rate regime. However, the model size remains an important issue in current state-of-the-art proposals and existing solutions require significant computation effort on the decoding side. This limits their usage in realistic scenarios and the extension to video compression. In this paper, we demonstrate how to leverage knowledge distillation to obtain equally capable image decoders at a fraction of the original number of parameters. We investigate several aspects of our solution including sequence specialization with side information for image coding. Finally, we also show how to transfer the obtained benefits into the setting of video compression. Overall, this allows us to reduce the model size by a factor of 20 and to achieve 50% reduction in decoding time.

Chinese characters, because of their complex structure and a large number, lead to an extremely high cost of time for designers to design a complete set of characters. As a result, the dramatic growth of characters used in various fields such as culture and business has formed a strong contradiction between supply and demand with Chinese font design. Although most of the existing Chinese characters transformation models greatly alleviate the demand for character usage, the semantics of the generated characters cannot be guaranteed and the generation efficiency is low. At the same time, the models require large amounts of paired data for training, which requires a large amount of sample processing time.

To address the problems of existing methods, this paper proposes an unsupervised Chinese characters generation method based on generative adversarial networks, which fuses Style-Attentional Net to a skip-connected U-Net as a+GAN generator network architecture. It effectively and flexibly integrates local style patterns based on the semantic spatial distribution of content images while retaining feature information of different sizes. Our model generates fonts that maintain the source domain content features and the target domain style features at the end of training. The addition of the style specification module and the classification discriminator allows the model to generate multiple style typefaces.

The generation results show that the model proposed in this paper can perform the task of Chinese character style-transfer well. The model generates high-quality images of Chinese characters and generates Chinese characters with complete structures and natural strokes.

In the quantitative comparison experiments and qualitative comparison experiments, our model has more superior visual effects and image performance indexes compared with the existing models. In sample size experiments, clearly structured fonts are still generated and the model demonstrates substantial robustness.

At the same time, the training conditions of our model are easy to meet and facilitate generalization to real applications.

[Keywords: Chinese characters, style transfer, generative adversarial networks, unsupervised learning, style-attentional networks]

Existing conditional image synthesis frameworks generate images based on user inputs in a single modality, such as text, segmentation, sketch, or style reference. They are often unable to leverage multimodal user inputs when available, which reduces their practicality. To address this limitation, we propose the Product-of-Experts Generative Adversarial Networks (PoE-GAN) framework, which can synthesize images conditioned on multiple input modalities or any subset of them, even the empty set. PoE-GAN consists of a product-of-experts generator and a multimodal multiscale projection discriminator. Through our carefully designed training scheme, PoE-GAN learns to synthesize images with high quality and diversity. Besides advancing the state of the art in multimodal conditional image synthesis, PoE-GAN also outperforms the best existing unimodal conditional image synthesis approaches when tested in the unimodal setting. The project website is available at https: /  ​ /  ​deepimagination.cc /  ​PoE-GAN /  ​⁠.

Generation of photo-realistic images, semantic editing and representation learning are a few of many potential applications of high resolution generative models.

Recent progress in GANs have established them as an excellent choice for such tasks. However, since they do not provide an inference model, image editing or downstream tasks such as classification can not be done on real images using the GAN latent space. Despite numerous efforts to train an inference model or design an iterative method to invert a pre-trained generator, previous methods are dataset (eg. human face images) and architecture (eg. StyleGAN) specific. These methods are nontrivial to extend to novel datasets or architectures.

We propose a general framework that is agnostic to architecture and datasets. Our key insight is that, by training the inference and the generative model together, we allow them to adapt to each other and to converge to a better quality model. Our InvGAN, short for Invertable GAN, successfully embeds real images to the latent space of a high quality generative model. This allows us to perform image inpainting, merging, interpolation and online data augmentation.

We demonstrate this with extensive qualitative and quantitative experiments.

Generating images from natural language instructions is an intriguing yet highly challenging task. We approach text-to-image generation by combining the power of the retrained CLIP representation with an off-the-shelf image generator (GANs), optimizing in the latent space of GAN to find images that achieve maximum CLIP score with the given input text. Compared to traditional methods that train generative models from text to image starting from scratch, the CLIP+GAN approach is training-free, zero shot and can be easily customized with different generators.

However, optimizing CLIP score in the GAN space casts a highly challenging optimization problem and off-the-shelf optimizers such as Adam fail to yield satisfying results. In this work, we propose a FuseDream pipeline, which improves the CLIP+GAN approach with three key techniques: (1) an AugCLIP score which robustifies the CLIP objective by introducing random augmentation on image. (2) a novel initialization and over-parameterization strategy for optimization which allows us to efficiently navigate the non-convex landscape in GAN space. (3) a composed generation technique which, by leveraging a novel bi-level optimization formulation, can compose multiple images to extend the GAN space and overcome the data-bias.

When promoted by different input text, FuseDream can generate high-quality images with varying objects, backgrounds, artistic styles, even novel counterfactual concepts that do not appear in the training data of the GAN we use. Quantitatively, the images generated by FuseDream yield top-level Inception score and FID score on MS COCO dataset, without additional architecture design or training.

Our code is publicly available at GitHub⁠.

Deep neural networks are vulnerable to attacks from adversarial inputs and, more recently, Trojans to misguide or hijack the decision of the model. We expose the existence of an intriguing class of bounded adversarial examples—Universal NaTuralistic adversarial paTches—we call TnTs, by exploring the superset of the bounded adversarial example space and the natural input space within generative adversarial networks. Now, an adversary can arm themselves with a patch that is naturalistic, less malicious-looking, physically realizable, highly effective—achieving high attack success rates, and universal. A TnT is universal because any input image captured with a TnT in the scene will: (1) misguide a network (untargeted attack); or (2) force the network to make a malicious decision (targeted attack). Interestingly, now, an adversarial patch attacker has the potential to exert a greater level of control—the ability to choose a location independent, natural-looking patch as a trigger in contrast to being constrained to noisy perturbations—an ability is thus far shown to be only possible with Trojan attack methods needing to interfere with the model building processes to embed a backdoor at the risk discovery; but, still realize a patch deployable in the physical world. Through extensive experiments on the large-scale visual classification task, ImageNet with evaluations across its entire validation set of 50,000 images, we demonstrate the realistic threat from TnTs and the robustness of the attack. We show a generalization of the attack to create patches achieving higher attack success rates than existing state-of-the-art methods. Our results show the generalizability of the attack to different visual classification tasks (CIFAR-10, GTSRB, PubFig) and multiple state-of-the-art deep neural networks such as WideResnet50, Inception-V3 and VGG-16⁠.

We introduce the GANformer2 model, an iterative object-oriented transformer, explored for the task of generative modeling. The network incorporates strong and explicit structural priors⁠, to reflect the compositional nature of visual scenes, and synthesizes images through a sequential process. It operates in two stages: a fast and lightweight planning phase, where we draft a high-level scene layout, followed by an attention-based execution phase, where the layout is being refined, evolving into a rich and detailed picture.

Our model moves away from conventional black-box GAN architectures that feature a flat and monolithic latent space towards a transparent design that encourages efficiency, controllability and interpretability. We demonstrate GANformer2’s strengths and qualities through a careful evaluation over a range of datasets, from multi-object CLEVR scenes to the challenging COCO images, showing it successfully achieves state-of-the-art performance in terms of visual quality, diversity and consistency.

Further experiments demonstrate the model’s disentanglement and provide a deeper insight into its generative process, as it proceeds step-by-step from a rough initial sketch, to a detailed layout that accounts for objects’ depths and dependencies, and up to the final high-resolution depiction of vibrant and intricate real-world scenes. See https:/​/​github.com/​dorarad/​gansformer for model implementation.

A wide variety of deep generative models has been developed in the past decade. Yet, these models often struggle with simultaneously addressing three key requirements including: high sample quality, mode coverage, and fast sampling. We call the challenge imposed by these requirements the generative learning trilemma, as the existing models often trade some of them for others. Particularly, denoising diffusion models have shown impressive sample quality and diversity, but their expensive sampling does not yet allow them to be applied in many real-world applications. In this paper, we argue that slow sampling in these models is fundamentally attributed to the Gaussian assumption in the denoising step which is justified only for small step sizes. To enable denoising with large steps, and hence, to reduce the total number of denoising steps, we propose to model the denoising distribution using a complex multimodal distribution. We introduce denoising diffusion generative adversarial networks (denoising diffusion GANs) that model each denoising step using a multimodal conditional GAN. Through extensive evaluations, we show that denoising diffusion GANs obtain sample quality and diversity competitive with original diffusion models while being 2000× faster on the CIFAR-10 dataset. Compared to traditional GANs, our model exhibits better mode coverage and sample diversity. To the best of our knowledge, denoising diffusion GAN is the first model that reduces sampling cost in diffusion models to an extent that allows them to be applied to real-world applications inexpensively.

Generative adversarial networks (GANs) have recently found applications in image editing. However, most GAN based image editing methods often require large scale datasets with semantic segmentation annotations for training, only provide high level control, or merely interpolate between different images. Here, we propose EditGAN, a novel method for high quality, high precision semantic image editing, allowing users to edit images by modifying their highly detailed part segmentation masks, eg. drawing a new mask for the headlight of a car. EditGAN builds on a GAN framework that jointly models images and their semantic segmentations, requiring only a handful of labeled examples, making it a scalable tool for editing. Specifically, we embed an image into the GAN latent space and perform conditional latent code optimization according to the segmentation edit, which effectively also modifies the image. To amortize optimization, we find editing vectors in latent space that realize the edits. The framework allows us to learn an arbitrary number of editing vectors, which can then be directly applied on other images at interactive rates. We experimentally show that EditGAN can manipulate images with an unprecedented level of detail and freedom, while preserving full image quality.We can also easily combine multiple edits and perform plausible edits beyond EditGAN training data. We demonstrate EditGAN on a wide variety of image types and quantitatively outperform several previous editing methods on standard editing benchmark tasks.

Generative Adversarial Networks (GANs) produce high-quality images but are challenging to train. They need careful regularization, vast amounts of compute, and expensive hyper-parameter sweeps. We make significant headway on these issues by projecting generated and real samples into a fixed, pretrained feature space. Motivated by the finding that the discriminator cannot fully exploit features from deeper layers of the pretrained model, we propose a more effective strategy that mixes features across channels and resolutions. Our Projected GAN improves image quality, sample efficiency, and convergence speed. It is further compatible with resolutions of up to one Megapixel and advances the state-of-the-art Fréchet Inception Distance (FID) on twenty-two benchmark datasets. Importantly, Projected GANs match the previously lowest FIDs up to 40× faster, cutting the wall-clock time from 5 days to less than 3 hours given the same computational resources.

Recent advances in efficient Transformers have exploited either the sparsity or low-rank properties of attention matrices to reduce the computational and memory bottlenecks of modeling long sequences. However, it is still challenging to balance the trade-off between model quality and efficiency to perform an one-size-fits-all approximation for different tasks. To better understand this trade-off, we observe that sparse and low-rank approximations excel in different regimes, determined by the softmax temperature in attention, and sparse + low-rank can outperform each individually. Inspired by the classical robust-PCA algorithm for sparse and low-rank decomposition, we propose Scatterbrain, a novel way to unify sparse (via locality sensitive hashing) and low-rank (via kernel feature map) attention for accurate and efficient approximation. The estimation is unbiased with provably low error. We empirically show that Scatterbrain can achieve 2.1× lower error than baselines when serving as a drop-in replacement in BigGAN image generation and pre-trained T2T-ViT⁠. On a pre-trained T2T Vision transformer, even without fine-tuning, Scatterbrain can reduce 98% of attention memory at the cost of only 1% drop in accuracy. We demonstrate Scatterbrain for end-to-end training with up to 4 points better perplexity and 5 points better average accuracy than sparse or low-rank efficient transformers on language modeling and long-range-arena tasks.

Transformer becomes prevalent in computer vision, especially for high-level vision tasks. However, deploying Transformer in the generative adversarial network (GAN) framework is still an open yet challenging problem. In this paper, we conduct a comprehensive empirical study to investigate the intrinsic properties of Transformer in GAN for high-fidelity image synthesis. Our analysis highlights the importance of feature locality in image generation. We first investigate the effective ways to implement local attention. We then examine the influence of residual connections in self-attention layers and propose a novel way to reduce their negative impacts on learning discriminators and conditional generators. Our study leads to a new design of Transformers in GAN, a convolutional neural network (CNN)-free generator termed as STrans-G, which achieves competitive results in both unconditional and conditional image generations. The Transformer-based discriminator, STrans-D, also significantly reduces its gap against the CNN-based discriminators.

Recently, Vision Transformers (ViTs) have shown competitive performance on image recognition while requiring less vision-specific inductive biases. In this paper, we investigate if such observation can be extended to image generation. To this end, we integrate the ViT architecture into generative adversarial networks (GANs). We observe that existing regularization methods for GANs interact poorly with self-attention, causing serious instability during training. To resolve this issue, we introduce novel regularization techniques for training GANs with ViTs. Empirically, our approach, named ViTGAN, achieves comparable performance to state-of-the-art CNN-based StyleGAN2 on CIFAR-10, CelebA⁠, and LSUN bedroom datasets.

Recent developments in deep learning have significantly improved the quality of synthesized singing voice audio. However, prominent neural singing voice synthesis systems suffer from slow inference speed due to their autoregressive design. Inspired by MLP-Mixer⁠, a novel architecture introduced in the vision literature for attention-free image classification, we propose MLP Singer, a parallel Korean singing voice synthesis system. To the best of our knowledge, this is the first work that uses an entirely MLP-based architecture for voice synthesis. Listening tests demonstrate that MLP Singer outperforms a larger autoregressive GAN-based system, both in terms of audio quality and synthesis speed. In particular, MLP Singer achieves a real-time factor of up to 200 and 3400 on CPUs and GPUs respectively, enabling order of magnitude faster generation on both environments.

We show how to learn a map that takes a content code, derived from a face image, and a randomly chosen style code to an anime image. We derive an adversarial loss from our simple and effective definitions of style and content. This adversarial loss guarantees the map is diverse—a very wide range of anime can be produced from a single content code. Under plausible assumptions, the map is not just diverse, but also correctly represents the probability of an anime, conditioned on an input face. In contrast, current multimodal generation procedures cannot capture the complex styles that appear in anime. Extensive quantitative experiments support the idea the map is correct. Extensive qualitative results show that the method can generate a much more diverse range of styles than SOTA comparisons. Finally, we show that our formalization of content and style allows us to perform video to video translation without ever training on videos.

[Paper⁠; blog] We show that cascaded diffusion models are capable of generating high fidelity images on the class-conditional ImageNet generation challenge, without any assistance from auxiliary image classifiers to boost sample quality.

A cascaded diffusion model comprises a pipeline of multiple diffusion models that generate images of increasing resolution, beginning with a standard diffusion model at the lowest resolution, followed by one or more super-resolution diffusion models that successively upsample the image and add higher resolution details.

We find that the sample quality of a cascading pipeline relies crucially on conditioning augmentation, our proposed method of data augmentation of the lower resolution conditioning inputs to the super-resolution models.

Our experiments show that conditioning augmentation prevents compounding error during sampling in a cascaded model, helping us to train cascading pipelines achieving FID scores of 1.48 at 64×64, 3.52 at 128×128 and 4.88 at 256×256 resolutions, outperforming BigGAN-deep, and classification accuracy scores of 63.02% (top-1) and 84.06% (top-5) at 256×256, outperforming VQ-VAE-2⁠.

• Cascaded Diffusion Models (CDM) are pipelines of diffusion models that generate images of increasing resolution.
• CDMs yield high fidelity samples superior to BigGAN-deep and VQ-VAE-2 in terms of both FID score and classification accuracy score on class-conditional ImageNet generation.
• These results are achieved with pure generative models without any classifier.
• We introduce conditioning augmentation, our data augmentation technique that we find critical towards achieving high sample fidelity.

[See also: “SR3: Image Super-Resolution via Iterative Refinement”⁠, Saharia et al 2021; “Diffusion Models Beat GANs on Image Synthesis”⁠, Dhariwal & Nichol 2021.]

…Concurrently, Dhariwal and Nichol showed that their diffusion models, named ADM, also outperform GANs on ImageNet generation. ADM achieves this result using classifier guidance, which boosts sample quality by modifying the diffusion sampling procedure to simultaneously maximize the score of an extra image classifier. As measured by FID score, ADM with classifier guidance outperforms our reported results, but our reported results outperform ADM without classifier guidance.

Our work is a demonstration of the effectiveness of pure generative models, namely cascaded diffusion models without the assistance of extra image classifiers. Nonetheless, classifier guidance and cascading are complementary techniques for improving sample quality, and a detailed investigation of how they interact is warranted.

While attention-based transformer networks achieve unparalleled success in nearly all language tasks, the large number of tokens coupled with the quadratic activation memory usage makes them prohibitive for visual tasks. As such, while language-to-language translation has been revolutionized by the transformer model, convolutional networks remain the de facto solution for image-to-image translation. The recently proposed MLP-Mixer architecture alleviates some of the speed and memory issues associated with attention-based networks while still retaining the long-range connections that make transformer models desirable. Leveraging this efficient alternative to self-attention, we propose a new unpaired image-to-image translation model called MixerGAN: a simpler MLP-based architecture that considers long-distance relationships between pixels without the need for expensive attention mechanisms. Quantitative and qualitative analysis shows that MixerGAN achieves competitive results when compared to prior convolutional-based methods.

We show that diffusion models can achieve image sample quality superior to the current state-of-the-art generative models.

We achieve this on unconditional image synthesis by finding a better architecture through a series of ablations. For conditional image synthesis, we further improve sample quality with classifier guidance: a simple, compute-efficient method for trading off diversity for sample quality using gradients from a classifier.

We achieve an FID of 2.97 on ImageNet 128×128, 4.59 on ImageNet 256×256, and 7.72 on ImageNet 512×512, and we match BigGAN-deep even with as few as 25 forward passes per sample, all while maintaining better coverage of the distribution. [nearest-neighbors]

Finally, we find that classifier guidance combines well with upsampling diffusion models, further improving FID to 3.85 on ImageNet 512×512.

We release our code [and checkpoints].

GANs largely increases the potential impact of generative models. Therefore, we propose a novel knowledge transfer method for generative models based on mining the knowledge that is most beneficial to a specific target domain, either from a single or multiple pretrained GANs. This is done using a miner network that identifies which part of the generative distribution of each pretrained GAN outputs samples closest to the target domain. Mining effectively steers GAN sampling towards suitable regions of the latent space, which facilitates the posterior finetuning and avoids pathologies of other methods, such as mode collapse and lack of flexibility. Furthermore, to prevent overfitting on small target domains, we introduce sparse subnetwork selection, that restricts the set of trainable neurons to those that are relevant for the target dataset. We perform comprehensive experiments on several challenging datasets using various GAN architectures (BigGAN⁠, Progressive GAN, and StyleGAN) and show that the proposed method, called MineGAN, effectively transfers knowledge to domains with few target images, outperforming existing methods. In addition, MineGAN can successfully transfer knowledge from multiple pretrained GANs.

Recent studies on Generative Adversarial Network (GAN) reveal that different layers of a generative CNN hold different semantics of the synthesized images. However, few GAN models have explicit dimensions to control the semantic attributes represented in a specific layer. This paper proposes EigenGAN which is able to unsupervisedly mine interpretable and controllable dimensions from different generator layers. Specifically, EigenGAN embeds one linear subspace with orthogonal basis into each generator layer. Via the adversarial training to learn a target distribution, these layer-wise subspaces automatically discover a set of “eigen-dimensions” at each layer corresponding to a set of semantic attributes or interpretable variations. By traversing the coefficient of a specific eigen-dimension, the generator can produce samples with continuous changes corresponding to a specific semantic attribute. Taking the human face for example, EigenGAN can discover controllable dimensions for high-level concepts such as pose and gender in the subspace of deep layers, as well as low-level concepts such as hue and color in the subspace of shallow layers. Moreover, under the linear circumstance, we theoretically prove that our algorithm derives the principal components as PCA does. Codes can be found in Github⁠.

[blog] We present SR3, an approach to image Super-Resolution via Repeated Refinement. SR3 adapts denoising diffusion probabilistic models to conditional image generation and performs super-resolution through a stochastic denoising process.

Inference starts with pure Gaussian noise and iteratively refines the noisy output using a U-Net model trained on denoising at various noise levels. SR3 exhibits strong performance on super-resolution tasks at different magnification factors, on faces and natural images.

We conduct human evaluation on a standard 8× face super-resolution task on CelebA-HQ, comparing with SOTA GAN methods. SR3 achieves a fool rate close to 50%, suggesting photo-realistic outputs, while GANs do not exceed a fool rate of 34%. We further show the effectiveness of SR3 in cascaded image generation, where generative models are chained with super-resolution models, yielding a competitive FID score of 11.3 on ImageNet.

This degree work shows the implementation of generative adversarial networks (GANs) to generate completely new illustrations, making use of comic images that have been previously studied, normalized, and filtered. The report will reflect the evolution of the research, which takes as a starting point the results of a previous research in which the first steps to follow in order to obtain a valid dataset are proposed.

The set of steps to obtain the final result are presented: the work methodology used, the remote work configuration using Google Colab⁠, the search and study of architectures and the evolution and improvement of the different networks to achieve the final results.

To carry out the generation of comic illustrations, a set of models is implemented progressively, taking an evolution from simpler to more complex and actual models, in order to better assimilate the necessary knowledge. In this way, the use of progressively growing generative adversarial networks or ProGANs is ultimately proposed. By using the ProGAN architecture, it is possible to improve the results compared to the use of traditional methods, obtaining images of great similarity to the original ones, maintaining the originality and avoiding the direct copy of images from the dataset.

In order to validate and demonstrate that the results obtained are replicable, a comparison between 2 different sets of images has been performed. Although both datasets have the same drawing style, they present substantial differences in their composition. On the one hand, results are shown for images of characters with the whole body and later using illustrations of character faces in the foreground.

Deep generative modelling is a class of techniques that train deep neural networks to model the distribution of training samples. Research has fragmented into various interconnected approaches, each of which making trade-offs including run-time, diversity, and architectural restrictions. In particular, this compendium covers energy-based models, variational autoencoders, generative adversarial networks, autoregressive models, normalizing flows, in addition to numerous hybrid approaches. These techniques are drawn under a single cohesive framework, comparing and contrasting to explain the premises behind each, while reviewing current state-of-the-art advances and implementations.

We introduce the GANformer, a novel and efficient type of transformer, and explore it for the task of visual generative modeling. The network employs a bipartite structure that enables long-range interactions across the image, while maintaining computation of linear efficiency, that can readily scale to high-resolution synthesis. It iteratively propagates information from a set of latent variables to the evolving visual features and vice versa, to support the refinement of each in light of the other and encourage the emergence of compositional representations of objects and scenes. In contrast to the classic transformer architecture, it utilizes multiplicative integration that allows flexible region-based modulation, and can thus be seen as a generalization of the successful StyleGAN network. We demonstrate the model’s strength and robustness through a careful evaluation over a range of datasets, from simulated multi-object environments to rich real-world indoor and outdoor scenes, showing it achieves state-of-the-art results in terms of image quality and diversity, while enjoying fast learning and better data-efficiency. Further qualitative and quantitative experiments offer us an insight into the model’s inner workings, revealing improved interpretability and stronger disentanglement, and illustrating the benefits and efficacy of our approach. An implementation of the model is available at Github⁠.

In this paper, we propose a novel framework to translate a portrait photo-face into an anime appearance. Our aim is to synthesize anime-faces which are style-consistent with a given reference anime-face. However, unlike typical translation tasks, such anime-face translation is challenging due to complex variations of appearances among anime-faces. Existing methods often fail to transfer the styles of reference anime-faces, or introduce noticeable artifacts/​distortions in the local shapes of their generated faces.

We propose AniGAN, a novel GAN-based translator that synthesizes high-quality anime-faces. Specifically, a new generator architecture is proposed to simultaneously transfer color/​texture styles and transform local facial shapes into anime-like counterparts based on the style of a reference anime-face, while preserving the global structure of the source photo-face. We propose a double-branch discriminator to learn both domain-specific distributions and domain-shared distributions, helping generate visually pleasing anime-faces and effectively mitigate artifacts.

Extensive experiments qualitatively and quantitatively demonstrate the superiority of our method over state-of-the-art methods.

Denoising diffusion probabilistic models (DDPM) are a class of generative models which have recently been shown to produce excellent samples.

We show that with a few simple modifications, DDPMs can also achieve competitive log-likelihoods while maintaining high sample quality. Additionally, we find that learning variances of the reverse diffusion process allows sampling with an order of magnitude fewer forward passes with a negligible difference in sample quality, which is important for the practical deployment of these models. We additionally use precision and recall to compare how well DDPMs and GANs cover the target distribution.

Finally, we show that the sample quality and likelihood of these models scale smoothly with model capacity and training compute, making them easily scalable.

We release our code at Github⁠.

…Figure 10 shows how FID and NLL improve relative to theoretical training compute.³ The FID curve looks ~linear on a log-log plot, suggesting that FID scales according to a power law (plotted as the black dashed line). The NLL curve does not fit a power law as cleanly, suggesting that validation NLL scales in a less-favorable manner than FID. This could be caused by a variety of factors, such as (1) an unexpectedly high irreducible loss (Henighan et al 2020) for this type of diffusion model, or (2) the model overfitting to the training distribution. We also note that these models do not achieve optimal log-likelihoods in general because they were trained with our L_hybrid objective and not directly with L_vlb to keep both good log-likelihoods and sample quality.

[In light of Chinchilla & related scaling work, the ‘bending’ here more likely reflects suboptimal scaling of their width vs depth multiplier, learning rate schedule etc.]

The recent explosive interest on transformers has suggested their potential to become powerful “universal” models for computer vision tasks, such as classification, detection, and segmentation. However, how further transformers can go—are they ready to take some more notoriously difficult vision tasks, eg. generative adversarial networks (GANs)?

Driven by that curiosity, we conduct the first pilot study in building a GAN completely free of convolutions, using only pure transformer-based architectures. Our vanilla GAN architecture, dubbed TransGAN, consists of a memory-friendly transformer-based generator that progressively increases feature resolution while decreasing embedding dimension, and a patch-level discriminator that is also transformer-based.

We then demonstrate TransGAN to notably benefit from data augmentations (more than standard GANs), a multi-task co-training strategy for the generator, and a locally initialized self-attention that emphasizes the neighborhood smoothness of natural images. Equipped with those findings, TransGAN can effectively scale up with bigger models and high-resolution image datasets.

Specifically, our best architecture achieves highly competitive performance compared to current state-of-the-art GANs based on convolutional backbones. Specifically, TransGAN sets new state-of-the-art IS score of 10.10 and FID score of 25.32 on STL-10⁠. It also reaches competitive 8.64 IS score and 11.89 FID score on CIFAR-10, and 12.23 FID score on CelebA 64×64, respectively.

We also conclude with a discussion of the current limitations and future potential of TransGAN. The code is available at https://github.com/VITA-Group/TransGAN⁠.

While we instantaneously recognize a face as attractive, it is much harder to explain what exactly defines personal attraction. This suggests that attraction depends on implicit processing of complex, culturally and individually defined features. Generative adversarial neural networks (GANs), which learn to mimic complex data distributions, can potentially model subjective preferences unconstrained by pre-defined model parameterization.

Here, we present generative brain-computer interfaces (GBCI), coupling GANs with brain-computer interfaces. GBCI first presents a selection of images and captures personalized attractiveness reactions toward the images via electroencephalography. These reactions are then used to control a ProGAN model, finding a representation that matches the features constituting an attractive image for an individual. We conducted an experiment (N = 30) to validate GBCI using a face-generating GAN and producing images that are hypothesized to be individually attractive. In double-blind evaluation of the GBCI-produced images against matched controls, we found GBCI yielded highly accurate results.

Thus, the use of EEG responses to control a GAN presents a valid tool for interactive information-generation. Furthermore, the GBCI-derived images visually replicated known effects from social neuroscience, suggesting that the individually responsive, generative nature of GBCI provides a powerful, new tool in mapping individual differences and visualizing cognitive-affective processing.

[Keywords: brain-computer interfaces, electroencephalography (EEG), generative adversarial networks (GANs), image generation, attraction, personal preferences, individual differences]

…Thus, negative generated images were evaluated as highly attractive for other people, but not for the participant themselves. Taken together, the results suggest that the GBCI was highly accurate in generating personally attractive images (83.33%). They also show that while both negative and positive generated images were evaluated as highly attractive for the general population (respectively M = 4.43 and 4.90 on a scale of 1–5), only the positive generated images (M = 4.57) were evaluated as highly personally attractive.

Qualitative results: In semi-structured post-test interviews, participants were shown the generated images that were expected to be found attractive/ unattractive. Thematic analysis found predictions of positive attractiveness were experienced as accurate: There were no false positives (generated unattractive found personally attractive). The participants also expressed being pleased with results (eg. “Quite an ideal beauty for a male!”; “I would be really attracted to this!”; “Can I have a copy of this? It looks just like my girlfriend!”).

In this research work we present CLIP-GLaSS, a novel zero-shot framework to generate an image (or a caption) corresponding to a given caption (or image). CLIP-GLaSS is based on the CLIP neural network, which, given an image and a descriptive caption, provides similar embeddings. Differently, CLIP-GLaSS takes a caption (or an image) as an input, and generates the image (or the caption) whose CLIP embedding is the most similar to the input one. This optimal image (or caption) is produced via a generative network, after an exploration by a genetic algorithm. Promising results are shown, based on the experimentation of the image Generators BigGAN and StyleGAN2, and of the text Generator GPT2

[Website demonstrating samples from a modified StyleGAN2 trained on Danbooru2019 using TRC TPUs for ~5m iterations for ~2 months on a TPUv3-32 pod; this modified ‘StyleGAN2-ext’, removes various regularizations which make StyleGAN2 data-efficient on datasets like FFHQ, but hobble its ability to model complicated images, and scales the model up >2×. This is surprisingly effective given StyleGAN’s previous inability to approach BigGAN’s Danbooru2019, and TADNE shows off the entertaining results.

The interface reuses Said Achmiz’s These Waifus Do Not Exist grid UI.

Writeup⁠; see also: Colab notebook to search by CLIP embedding; “This Waifu Does Not Exist” (TWDNE)⁠/​“This Fursona Does Not Exist” (TFDNE)⁠/​“This Pony Does Not Exist” (TPDNE)⁠, TADNE face editing⁠, CLIP=guided ponies]

[blog] The output of text-to-image synthesis systems should be coherent, clear, photo-realistic scenes with high semantic fidelity to their conditioned text descriptions. Our Cross-Modal Contrastive Generative Adversarial Network (XMC-GAN) addresses this challenge by maximizing the mutual information between image and text.

It does this via multiple contrastive losses which capture inter-modality and intra-modality correspondences. XMC-GAN uses an attentional self-modulation generator, which enforces strong text-image correspondence, and a contrastive discriminator, which acts as a critic as well as a feature encoder for contrastive learning.

The quality of XMC-GAN’s output is a major step up from previous models, as we show on three challenging datasets. On MS-COCO⁠, not only does XMC-GAN improve state-of-the-art FID from 24.70 to 9.33, but—more importantly—people prefer XMC-GAN by 77.3 for image quality and 74.1 for image-text alignment, compared to three other recent models. XMC-GAN also generalizes to the challenging Localized Narratives dataset (which has longer, more detailed descriptions), improving state-of-the-art FID from 48.70 to 14.12. Lastly, we train and evaluate XMC-GAN on the challenging Open Images data, establishing a strong benchmark FID score of 26.91.

We address a novel problem of stylized-colorization which colorizes a given line art using a given coloring style in text.

This problem can be stated as multi-domain image translation and is more challenging than the current colorization problem because it requires not only capturing the illustration distribution but also satisfying the required coloring styles specific to anime such as lightness, shading, or saturation.

We propose a GAN-based end-to-end model for stylized-colorization where the model has one generator and two discriminators. Our generator is based on the U-Net architecture and receives a pair of a line art and a coloring style in text as its input to produce a stylized-colorization image of the line art. Two discriminators, on the other hand, share weights at early layers to judge the stylized-colorization image in two different aspects: one for color and one for style. One generator and two discriminators are jointly trained in an adversarial and end-to-end manner.

Extensive experiments demonstrate the effectiveness of our proposed model.

[CLIP paper] We present a neural network that aims to address these problems: it is trained on a wide variety of images with a wide variety of natural language supervision that’s abundantly available on the internet. By design, the network can be instructed in natural language to perform a great variety of classification benchmarks, without directly optimizing for the benchmark’s performance, similar to the “zero-shot” capabilities of GPT-2 and GPT-3⁠. This is a key change: by not directly optimizing for the benchmark, we show that it becomes much more representative: our system closes this “robustness gap” by up to 75% while matching the performance of the original ResNet50 on ImageNet zero-shot without using any of the original 1.28M labeled examples.

Approach: We show that scaling a simple pre-training task is sufficient to achieve competitive zero-shot performance on a great variety of image classification datasets. Our method uses an abundantly available source of supervision: the text paired with images found across the internet. This data is used to create the following proxy training task for CLIP: given an image, predict which out of a set of 32,768 randomly sampled text snippets, was actually paired with it in our dataset.

In order to solve this task, our intuition is that CLIP models will need to learn to recognize a wide variety of visual concepts in images and associate them with their names. As a result, CLIP models can then be applied to nearly arbitrary visual classification tasks. For instance, if the task of a dataset is classifying photos of dogs vs cats we check for each image whether a CLIP model predicts the text description “a photo of a dog” or “a photo of a cat” is more likely to be paired with it.

1. CLIP is highly efficient…In the end, our best performing CLIP model trains on 256 GPUs for 2 weeks which is similar to existing large scale image models.
2. CLIP is flexible and general: Because they learn a wide range of visual concepts directly from natural language, CLIP models are substantially more flexible and general than existing ImageNet models. We find they are able to zero-shot perform many different tasks. To validate this we have measured CLIP’s zero-shot performance on over 30 different datasets including tasks such as fine-grained object classification, geo-localization, action recognition in videos, and OCR. [While CLIP’s zero-shot OCR performance is mixed, its semantic OCR representation is quite useful. When evaluated on the SST-2 NLP dataset rendered as images, a linear classifier on CLIP’s representation matches a CBoW model with direct access to the text. CLIP is also competitive at detecting hateful memes without needing ground truth text.] In particular, learning OCR is an example of an exciting behavior that does not occur in standard ImageNet models.

…CLIP allows people to design their own classifiers and removes the need for task-specific training data. [See also “AudioCLIP: Extending CLIP to Image, Text and Audio”⁠, Guzhov et al 2021; CLIP notebook compilation for art⁠, “Alien Dreams: An Emerging Art Scene”⁠/​“AI Generated Art Scene Explodes as Hackers Create Groundbreaking New Tools”⁠.]

Designed to learn long-range interactions on sequential data, transformers continue to show state-of-the-art results on a wide variety of tasks. In contrast to CNNs, they contain no inductive bias that prioritizes local interactions. This makes them expressive, but also computationally infeasible for long sequences, such as high-resolution images. We demonstrate how combining the effectiveness of the inductive bias of CNNs with the expressivity of transformers enables them to model and thereby synthesize high-resolution images. We show how to (1) use CNNs to learn a context-rich vocabulary of image constituents, and in turn (2) utilize transformers to efficiently model their composition within high-resolution images. Our approach is readily applied to conditional synthesis tasks, where both non-spatial information, such as object classes, and spatial information, such as segmentations, can control the generated image. In particular, we present the first results on semantically-guided synthesis of megapixel images with transformers. [Github⁠; Arxiv]

Recent advances in generative adversarial networks (GANs) have shown remarkable progress in generating high-quality images. However, this gain in performance depends on the availability of a large amount of training data. In limited data regimes, training typically diverges, and therefore the generated samples are of low quality and lack diversity. Previous works have addressed training in low data setting by leveraging transfer learning and data augmentation techniques.

We propose a novel transfer learning method for GANs in the limited data domain by leveraging informative data prior derived from self-supervised/​supervised pre-trained networks trained on a diverse source domain. We perform experiments on several standard vision datasets using various GAN architectures (BigGAN⁠, SNGAN, StyleGAN2) to demonstrate that the proposed method effectively transfers knowledge to domains with few target images, outperforming existing state-of-the-art techniques in terms of image quality and diversity. We also show the utility of data instance prior in large-scale unconditional image generation and image editing tasks.

Line-arts are used in many ways in the media industry. However, line-art colorization is tedious, labor-intensive, and time consuming. For such reasons, a Generative Adversarial Network (GAN)-based image-to-image colorization method has received much attention because of its promising results. In this paper, we propose to use color a point hinting method with two GAN-based generators used for enhancing the image quality. To improve the coloring performance of drawing with various line styles, generator takes account of the loss of the line-art. We propose a Line Detection Model (LDM) which is used in measuring line loss. LDM is a method of extracting line from a color image. We also propose histogram equalizer in the input line-art to generalize the distribution of line styles. This approach allows the generalization of the distribution of line style without increasing the complexity of inference stage. In addition, we propose seven segment hint pointing constraints to evaluate the colorization performance of the model with Fréchet Inception Distance (FID) score. We present visual and qualitative evaluations of the proposed methods. The result shows that using histogram equalization and LDM enabled line loss exhibits the best result. The Base model with XDoG (eXtended Difference-Of-Gaussians)generated line-art with and without color hints exhibits FID for colorized images score of 35.83 and 44.70, respectively, whereas the proposed model in the same scenario exhibits 32.16 and 39.77, respectively.

Despite the persistent effort by law enforcement, illicit drug trafficking in darknet markets has shown great resilience with new markets rapidly appearing after old ones being shut down. In order to more effectively detect, disrupt and dismantle illicit drug trades, there’s an imminent need to gain a deeper understanding toward the operations and dynamics of illicit drug trading activities. To address this challenge, in this paper, we design and develop an intelligent system (named dStyle-GAN) to automate the analysis for drug identification in darknet markets, by considering both content-based and style-aware information.

To determine whether a given pair of posted drugs are the same or not, in dStyle-GAN, based on the large-scale data collected from darknet markets, we first present an attributed heterogeneous information network (AHIN) to depict drugs, vendors, texts and writing styles, photos and photography styles, and the rich relations among them; and then we propose a novel generative adversarial network (GAN) based model over AHIN to capture the underlying distribution of posted drugs’ writing and photography styles to learn robust representations of drugs for their identifications. Unlike existing approaches, our proposed GAN-based model jointly considers the heterogeneity of network and relatedness over drugs formulated by domain-specific meta-paths for robust node (ie. drug) representation learning. To the best of our knowledge, the proposed dStyle-GAN represents the first principled GAN-based solution over graphs to simultaneously consider writing and photography styles as well as their latent distributions for node representation learning.

Extensive experimental results based on large-scale datasets collected from 6 darknet markets and the obtained ground-truth demonstrate that dStyle-GAN outperforms the state-of-the-art methods. Based on the identified drug pairs in the wild by dStyle-GAN, we perform further analysis to gain deeper insights into the dynamics and evolution of illicit drug trading activities in darknet markets, whose findings may facilitate law enforcement for proactive interventions.

In this paper, we use Generative Adversarial Networks (GAN) to address the industrial needs of auto colorization of line arts which takes enormous amount of manual labor. Auto-colorization method used in Image-to-Image conversion based on GAN has received a lot of attention due to its promising results.

In this paper, we present a solution to not only colorize the line art but also transform the low resolution out image to match the resolution of the input image through 2 generators and frequency separation method. A high frequency components are extracted from the line, then 2 generators are used to colorize the image in low resolution. The high frequency component is merged with low resolution image to produce the high resolution colorized image. The resolution of final output image matches the resolution of original image while preserving the texture of the input image, whereas the other schemes reduce the output image to 512 pixels.

We performed visual and qualitative evaluation using FID⁠, PSNR⁠, and SSIM⁠. The FID Score of the proposed method is better than the base model by about 4 (proposed: 47.87 and base model 51.64). PNSR and SSIM of the high-resolution images are also better than the base model. PSNR and SSIM of base model is 13.01 and 0.72 whereas the proposed is 20.77 and 0.86, respectively.

[Keywords: machine learning, Generative Adversarial Network, line arts colorization, image generation]

Current GAN-based art generation methods produce unoriginal artwork due to their dependence on conditional input. Here, we propose Sketch-And-Paint GAN (SAPGAN), the first model which generates Chinese landscape paintings from end to end, without conditional input. SAPGAN is composed of two GANs: SketchGAN for generation of edge maps, and PaintGAN for subsequent edge-to-painting translation. Our model is trained on a new dataset of traditional Chinese landscape paintings never before used for generative research. A 242-person Visual Turing Test study reveals that SAPGAN paintings are mistaken as human artwork with 55% frequency, significantly outperforming paintings from baseline GANs. Our work lays a groundwork for truly machine-original art generation.

Given the ever-increasing computational costs of modern machine learning models, we need to find new ways to reuse such expert models and thus tap into the resources that have been invested in their creation. Recent work suggests that the power of these massive models is captured by the representations they learn. Therefore, we seek a model that can relate between different existing representations and propose to solve this task with a conditionally invertible network. This network demonstrates its capability by (1) providing generic transfer between diverse domains, (2) enabling controlled content synthesis by allowing modification in other domains, and (3) facilitating diagnosis of existing representations by translating them into interpretable domains such as images. Our domain transfer network can translate between fixed representations without having to learn or finetune them. This allows users to utilize various existing domain-specific expert models from the literature that had been trained with extensive computational resources. Experiments on diverse conditional image synthesis tasks, competitive image modification results and experiments on image-to-image and text-to-image generation demonstrate the generic applicability of our approach. For example, we translate between BERT and BigGAN⁠, state-of-the-art text and image models to provide text-to-image generation, which neither of both experts can perform on their own.

Generative Adversarial Networks (GANs) have shown remarkable performance in image synthesis tasks, but typically require a large number of training samples to achieve high-quality synthesis. This paper proposes a simple and effective method, Few-Shot GAN (FSGAN), for adapting GANs in few-shot settings (less than 100 images). FSGAN repurposes component analysis techniques and learns to adapt the singular values of the pre-trained weights while freezing the corresponding singular vectors. This provides a highly expressive parameter space for adaptation while constraining changes to the pretrained weights. We validate our method in a challenging few-shot setting of 5–100 images in the target domain. We show that our method has significant visual quality gains compared with existing GAN adaptation methods. We report qualitative and quantitative results showing the effectiveness of our method. We additionally highlight a problem for few-shot synthesis in the standard quantitative metric used by data-efficient image synthesis works. Code and additional results are available at https://github.com/e-271/few-shot-gan .

We propose a novel type of balanced clustering algorithm to approximate attention. Attention complexity is reduced from \U0001D4AA(N²) to \U0001D4AA(N log N), where N is the sequence length.

Our algorithm, SMYRF, uses Locality Sensitive Hashing (LSH) in a novel way by defining new Asymmetric transformations and an adaptive scheme that produces balanced clusters. The biggest advantage of SMYRF is that it can be used as a drop-in replacement for dense attention layers without any retraining. On the contrary, prior fast attention methods impose constraints (eg. queries and keys share the same vector representations) and require re-training from scratch.

We apply our method to pre-trained state-of-the-art Natural Language Processing and Computer Vision models and we report substantial memory and speed benefits. Notably, SMYRF-BERT outperforms (slightly) BERT on GLUE⁠, while using 50% less memory. We also show that SMYRF can be used interchangeably with dense attention before and after training. Finally, we use SMYRF to train GANs with attention in high resolutions. Using a single TPU, we were able to scale attention to 128×128 = 16k and 256×256 = 65k tokens on BigGAN on CelebA-HQ.

Image and video synthesis are closely related areas aiming at generating content from noise. While rapid progress has been demonstrated in improving image-based models to handle large resolutions, high-quality renderings, and wide variations in image content, achieving comparable video generation results remains problematic.

We present a framework that leverages contemporary image generators to render high-resolution videos. We frame the video synthesis problem as discovering a trajectory in the latent space of a pre-trained and fixed image generator. Not only does such a framework render high-resolution videos, but it also is an order of magnitude more computationally efficient.

We introduce a motion generator that discovers the desired trajectory, in which content and motion are disentangled. With such a representation, our framework allows for a broad range of applications, including content and motion manipulation. Furthermore, we introduce a new task, which we call cross-domain video synthesis, in which the image and motion generators are trained on disjoint datasets belonging to different domains. This allows for generating moving objects for which the desired video data is not available.

Extensive experiments on various datasets demonstrate the advantages of our methods over existing video generation techniques. Code will be released⁠.

[Keywords: high-resolution video generation, contrastive learning, cross-domain video generation]

A computational-efficient GAN for few-shot hi-fi image dataset (converge on single GPU with few hours’ training, on 1024 resolution sub-hundred images).

Training Generative Adversarial Networks (GAN) on high-fidelity images usually requires large-scale GPU-clusters and a vast number of training images. In this paper, we study the few-shot image synthesis task for GAN with minimum computing cost. We propose a light-weight GAN structure that gains superior quality on 1024×1024px resolution. Notably, the model converges from scratch with just a few hours of training on a single RTX-2080 GPU; and has a consistent performance, even with less than 100 training samples. 2 technique designs constitute our work, a skip-layer channel-wise excitation module and a self-supervised discriminator trained as a feature-encoder. With 13 datasets covering a wide variety of image domains, we show our model’s robustness and its superior performance compared to the state-of-the-art StyleGAN2⁠.

[Keywords: deep learning, generative model, image synthesis, few-shot learning, generative adversarial network, self-supervised learning, unsupervised learning]

State-of-the-art models for high-resolution image generation, such as BigGAN and VQVAE-2, require an incredible amount of compute resources and/​or time (512 TPU-v3 cores) to train, putting them out of reach for the larger research community. On the other hand, GAN-based image super-resolution models, such as ESRGAN, can not only upscale images to high dimensions, but also are efficient to train. In this paper, we present not-so-big-GAN (nsb-GAN), a simple yet cost-effective two-step training framework for deep generative models (DGMs) of high-dimensional natural images. First, we generate images in low-frequency bands by training a sampler in the wavelet domain. Then, we super-resolve these images from the wavelet domain back to the pixel-space with our novel wavelet super-resolution decoder network. Wavelet-based down-sampling method preserves more structural information than pixel-based methods, leading to significantly better generative quality of the low-resolution sampler (eg. 64×64). Since the sampler and decoder can be trained in parallel and operate on much lower dimensional spaces than end-to-end models, the training cost is substantially reduced. On ImageNet 512×512, our model achieves a Fréchet Inception Distance (FID) of 10.59—beating the baseline BigGAN model—at half the compute (256 TPU-v3 cores).

[blog] Spectral normalization (SN) is a widely-used technique for improving the stability and sample quality of Generative Adversarial Networks (GANs). However, there is currently limited understanding of why SN is effective. In this work, we show that SN controls two important failure modes of GAN training: exploding and vanishing gradients. Our proofs illustrate a (perhaps unintentional) connection with the successful LeCun initialization. This connection helps to explain why the most popular implementation of SN for GANs requires no hyper-parameter tuning, whereas stricter implementations of SN have poor empirical performance out-of-the-box. Unlike LeCun initialization which only controls gradient vanishing at the beginning of training, SN preserves this property throughout training. Building on this theoretical understanding, we propose a new spectral normalization technique: Bidirectional Scaled Spectral Normalization (BSSN), which incorporates insights from later improvements to LeCun initialization: Xavier initialization and Kaiming initialization. Theoretically, we show that BSSN gives better gradient control than SN. Empirically, we demonstrate that it outperforms SN in sample quality and training stability on several benchmark datasets.

A deep generative model such as a GAN learns to model a rich set of semantic and physical rules about the target distribution, but up to now, it has been obscure how such rules are encoded in the network, or how a rule could be changed. In this paper, we introduce a new problem setting: manipulation of specific rules encoded by a deep generative model. To address the problem, we propose a formulation in which the desired rule is changed by manipulating a layer of a deep network as a linear associative memory. We derive an algorithm for modifying one entry of the associative memory, and we demonstrate that several interesting structural rules can be located and modified within the layers of state-of-the-art generative models. We present an user interface to enable users to interactively change the rules of a generative model to achieve desired effects, and we show several proof-of-concept applications. Finally, results on multiple datasets demonstrate the advantage of our method against standard fine-tuning methods and edit transfer algorithms.

We present high quality image synthesis results using diffusion probabilistic models, a class of latent variable models inspired by considerations from nonequilibrium thermodynamics.

Our best results are obtained by training on a weighted variational bound designed according to a novel connection between diffusion probabilistic models and denoising score matching with Langevin dynamics, and our models naturally admit a progressive lossy decompression scheme that can be interpreted as a generalization of autoregressive decoding.

On the unconditional CIFAR10 dataset, we obtain an Inception score of 9.46 and a state-of-the-art FID score of 3.17. On 256×256 LSUN, we obtain sample quality similar to Progressive GAN. Our implementation is available at Github⁠.

The performance of generative adversarial networks (GANs) heavily deteriorates given a limited amount of training data. This is mainly because the discriminator is memorizing the exact training set. To combat it, we propose Differentiable Augmentation (DiffAugment), a simple method that improves the data efficiency of GANs by imposing various types of differentiable augmentations on both real and fake samples. Previous attempts to directly augment the training data manipulate the distribution of real images, yielding little benefit; DiffAugment enables us to adopt the differentiable augmentation for the generated samples, effectively stabilizes training, and leads to better convergence. Experiments demonstrate consistent gains of our method over a variety of GAN architectures and loss functions for both unconditional and class-conditional generation. With DiffAugment, we achieve a state-of-the-art FID of 6.80 with an IS of 100.8 on ImageNet 128×128 and 2–4× reductions of FID given 1,000 images on FFHQ and LSUN. Furthermore, with only 20% training data, we can match the top performance on CIFAR-10 and CIFAR-100. Finally, our method can generate high-fidelity images using only 100 images without pre-training, while being on par with existing transfer learning algorithms. Code is available⁠.

Despite success on a wide range of problems related to vision, generative adversarial networks (GANs) often suffer from inferior performance due to unstable training, especially for text generation.

To solve this issue, we propose a new variational GAN training framework which enjoys superior training stability. Our approach is inspired by a connection of GANs and reinforcement learning under a variational perspective. The connection leads to (1) probability ratio clipping that regularizes generator training to prevent excessively large updates, and (2) a sample re-weighting mechanism that improves discriminator training by downplaying bad-quality fake samples. Moreover, our variational GAN framework can provably overcome the training issue in many GANs that an optimal discriminator cannot provide any informative gradient to training generator.

By plugging the training approach in diverse state-of-the-art GAN architectures, we obtain substantially improved performance over a range of tasks, including text generation, text style transfer, and image generation.

Recent successes in Generative Adversarial Networks (GAN) have affirmed the importance of using more data in GAN training. Yet it is expensive to collect data in many domains such as medical applications. Data Augmentation (DA) has been applied in these applications.

In this work, we first argue that the classical DA approach could mislead the generator to learn the distribution of the augmented data, which could be different from that of the original data. We then propose a principled framework, termed Data Augmentation Optimized for GAN (DAG), to enable the use of augmented data in GAN training to improve the learning of the original distribution. We provide theoretical analysis to show that using our proposed DAG aligns with the original GAN in minimizing the Jensen-Shannon (JS) divergence between the original distribution and model distribution. Importantly, the proposed DAG effectively leverages the augmented data to improve the learning of discriminator and generator.

We conduct experiments to apply DAG to different GAN models: unconditional GAN, conditional GAN, self-supervised GAN and CycleGAN using datasets of natural images and medical images. The results show that DAG achieves consistent and considerable improvements across these models. Furthermore, when DAG is used in some GAN models, the system establishes state-of-the-art Fréchet Inception Distance (FID) scores.

Our code is available.

Data augmentations have been widely studied to improve the accuracy and robustness of classifiers. However, the potential of image augmentation in improving GAN models for image synthesis has not been thoroughly investigated in previous studies.

In this work, we systematically study the effectiveness of various existing augmentation techniques for GAN training in a variety of settings. We provide insights and guidelines on how to augment images for both vanilla GANs and GANs with regularizations, improving the fidelity of the generated images substantially. Surprisingly, we find that vanilla GANs attain generation quality on par with recent state-of-the-art results if we use augmentations on both real and generated images. When this GAN training is combined with other augmentation-based regularization techniques, such as contrastive loss and consistency regularization, the augmentations further improve the quality of generated images.

We provide new state-of-the-art results for conditional generation on CIFAR-10 with both consistency loss and contrastive loss as additional regularizations.

Simulation is a crucial component of any robotic system. In order to simulate correctly, we need to write complex rules of the environment: how dynamic agents behave, and how the actions of each of the agents affect the behavior of others. In this paper, we aim to learn a simulator by simply watching an agent interact with an environment. We focus on graphics games as a proxy of the real environment. We introduce GameGAN, a generative model that learns to visually imitate a desired game by ingesting screenplay and keyboard actions during training. Given a key pressed by the agent, GameGAN “renders” the next screen using a carefully designed generative adversarial network. Our approach offers key advantages over existing work: we design a memory module that builds an internal map of the environment, allowing for the agent to return to previously visited locations with high visual consistency. In addition, GameGAN is able to disentangle static and dynamic components within an image making the behavior of the model more interpretable, and relevant for downstream tasks that require explicit reasoning over dynamic elements. This enables many interesting applications such as swapping different components of the game to build new games that do not exist.

Description of 3 anime datasets for machine learning based on Danbooru: cropped anime faces, whole-single-character crops, and hand crops (with hand detection model).

Documentation of 3 anime datasets for machine learning based on Danbooru: 300k cropped anime faces (primarily used for StyleGAN⁠/​This Waifu Does Not Exist), 855k whole-single-character figure crops (extracted from Danbooru using AniSeg), and 58k hand crops (based on a dataset of 14k hand-annotated bounding boxes used to train a YOLOv3 hand detection model).

These datasets can be used for machine learning directly, or included as data augmentation: faces, figures, and hands are some of the most noticeable features of anime images, and by cropping images down to just those 3 features, they can enhance modeling of those by eliminating distracting context, zooming in, and increasing the weight during training.

[Project page for GameGAN, a GAN which can learn and model a playable arcade game such as Pac-Man. This landing page includes diagrams, videos demonstrating GameGAN trained on the official version of Pac-Man, GameGAN trained on a custom version of Pac-Man and VizDoom, GameGAN trained with memory module & disentangling rendering engine, and swapping foreground / background.]

Human sketches can be expressive and abstract at the same time. Generating anime avatars from simple or even bad face drawing is an interesting area. Lots of related work has been done such as auto-coloring sketches to anime or transforming real photos to anime. However, there aren’t many interesting works yet to show how to generate anime avatars from just some simple drawing input.

In this project, we propose using GAN to generate anime avatars from sketches.

Normalization layers and activation functions are fundamental components in deep networks and typically co-locate with each other. Here we propose to design them using an automated approach. Instead of designing them separately, we unify them into a single tensor-to-tensor computation graph, and evolve its structure starting from basic mathematical functions. Examples of such mathematical functions are addition, multiplication and statistical moments. The use of low-level mathematical functions, in contrast to the use of high-level modules in mainstream NAS, leads to a highly sparse and large search space which can be challenging for search methods. To address the challenge, we develop efficient rejection protocols to quickly filter out candidate layers that do not work well. We also use multi-objective evolution to optimize each layer’s performance across many architectures to prevent overfitting.

Our method leads to the discovery of EvoNorms, a set of new normalization-activation layers with novel, and sometimes surprising structures that go beyond existing design patterns. For example, some EvoNorms do not assume that normalization and activation functions must be applied sequentially, nor need to center the feature maps, nor require explicit activation functions. Our experiments show that EvoNorms work well on image classification models including ResNets⁠, MobileNets and EfficientNets but also transfer well to Mask R-CNN with FPN/​SpineNet for instance segmentation and to BigGAN for image synthesis, outperforming BatchNorm and GroupNorm based layers in many cases.

This paper describes a simple technique to analyze Generative Adversarial Networks (GANs) and create interpretable controls for image synthesis, such as change of viewpoint, aging, lighting, and time of day. We identify important latent directions based on Principal Components Analysis (PCA) applied either in latent space or feature space. Then, we show that a large number of interpretable controls can be defined by layer-wise perturbation along the principal directions. Moreover, we show that BigGAN can be controlled with layer-wise inputs in a StyleGAN-like manner. We show results on different GANs trained on various datasets, and demonstrate good qualitative matches to edit directions found through earlier supervised approaches.

The primary aim of single-image super-resolution is to construct high-resolution (HR) images from corresponding low-resolution (LR) inputs. In previous approaches, which have generally been supervised, the training objective typically measures a pixel-wise average distance between the super-resolved (SR) and HR images. Optimizing such metrics often leads to blurring, especially in high variance (detailed) regions. We propose an alternative formulation of the super-resolution problem based on creating realistic SR images that downscale correctly. We present an algorithm addressing this problem, PULSE (Photo Upsampling via Latent Space Exploration), which generates high-resolution, realistic images at resolutions previously unseen in the literature. It accomplishes this in an entirely self-supervised fashion and is not confined to a specific degradation operator used during training, unlike previous methods (which require supervised training on databases of LR-HR image pairs). Instead of starting with the LR image and slowly adding detail, PULSE traverses the high-resolution natural image manifold, searching for images that downscale to the original LR image. This is formalized through the “downscaling loss”, which guides exploration through the latent space of a generative model. By leveraging properties of high-dimensional Gaussians, we restrict the search space to guarantee realistic outputs. PULSE thereby generates super-resolved images that both are realistic and downscale correctly. We show proof of concept of our approach in the domain of face super-resolution (ie. face hallucination). We also present a discussion of the limitations and biases of the method as currently implemented with an accompanying model card with relevant metrics. Our method outperforms state-of-the-art methods in perceptual quality at higher resolutions and scale factors than previously possible.

Among the major remaining challenges for generative adversarial networks (GANs) is the capacity to synthesize globally and locally coherent images with object shapes and textures indistinguishable from real images.

To target this issue we propose an alternative U-Net based discriminator architecture, borrowing the insights from the segmentation literature. The proposed U-Net based architecture allows to provide detailed per-pixel feedback to the generator while maintaining the global coherence of synthesized images, by providing the global image feedback as well. Empowered by the per-pixel response of the discriminator, we further propose a per-pixel consistency regularization technique based on the CutMix data augmentation, encouraging the U-Net discriminator to focus more on semantic and structural changes between real and fake images. This improves the U-Net discriminator training, further enhancing the quality of generated samples.

The novel discriminator improves over the state of the art in terms of the standard distribution and image quality metrics, enabling the generator to synthesize images with varying structure, appearance and levels of detail, maintaining global and local realism. Compared to the BigGAN baseline, we achieve an average improvement of 2.7 FID points across FFHQ, CelebA, and the newly introduced COCO-Animals dataset.

The code is available at Github⁠.

Tool for getting the dataset of cropped faces from [furry booru] e621 (NSFW; WikiFur description). It was created by training a YOLOv3 network on annotated facial features from about 1500 faces.

The total dataset includes ~186k faces. Rather than provide the cropped images, this repo contains CSV files with the bounding boxes of the detected features from my trained network, and a script to download the images from e621 and crop them based on these CSVs.

The CSVs also contain a subset of tags, which could potentially be used as labels to train a conditional GAN.

We introduce a simple (one line of code) modification to the Generative Adversarial Network (GAN) training algorithm that materially improves results with no increase in computational cost: When updating the generator parameters, we simply zero out the gradient contributions from the elements of the batch that the critic scores as ‘least realistic’. Through experiments on many different GAN variants, we show that this ‘top-k update’ procedure is a generally applicable improvement. In order to understand the nature of the improvement, we conduct extensive analysis on a simple mixture-of-Gaussians dataset and discover several interesting phenomena. Among these is that, when gradient updates are computed using the worst-scoring batch elements, samples can actually be pushed further away from their nearest mode. We also apply our method to recent GAN variants and improve state-of-the-art FID for conditional generation from 9.21 to 8.57 on CIFAR-10.

Recent work has increased the performance of Generative Adversarial Networks (GANs) by enforcing a consistency cost on the discriminator. We improve on this technique in several ways.

We first show that consistency regularization can introduce artifacts into the GAN samples and explain how to fix this issue. We then propose several modifications to the consistency regularization procedure designed to improve its performance.

We carry out extensive experiments quantifying the benefit of our improvements. For unconditional image synthesis on CIFAR-10 and CelebA, our modifications yield the best known FID scores on various GAN architectures. For conditional image synthesis on CIFAR-10, we improve the state-of-the-art FID score from 11.48 to 9.21. Finally, on ImageNet-2012, we apply our technique to the original BigGAN model and improve the FID from 6.66 to 5.38, which is the best score at that model size.

With the success of modern machine learning, it is becoming increasingly important to understand and control how learning algorithms interact. Unfortunately, negative results from game theory show there is little hope of understanding or controlling general n-player games. We therefore introduce smooth markets (SM-games), a class of n-player games with pairwise zero sum interactions. SM-games codify a common design pattern in machine learning that includes (some) GANs, adversarial training, and other recent algorithms. We show that SM-games are amenable to analysis and optimization using first-order methods.

Human facial expression transfer has been well explored using Generative Adversarial Networks. Also, in case of anime style images, several successful attempts have been made to generate high-quality anime face images using GAN approach. However, the task of anime facial expression transfer is not well studied yet due to the lack of a clean labeled anime dataset.

We address this issue from both data and model perspectives, by providing a clean labeled anime dataset and leveraging the use of the StarGAN image-to-image translation framework. Our collected dataset consists of about 5k high-quality anime face images including 5 major emotions collected from online image boards. We preprocessed our dataset by CARN super-resolution technique to improve quality of the images, and applied tuned StarGAN model to learn the mapping of an input anime image with arbitrary expression to the target expression.

We evaluate our work by visually comparing the output translated results with the baseline model. Moreover, we provide a quantitative analysis of our proposed approach by computing the confusion matrix of expression transfer accuracy.

[Keywords: facial expression transfer, unpaired image translation, Generative Adversarial Network, anime generation]

One of the attractive characteristics of deep neural networks is their ability to transfer knowledge obtained in one domain to other related domains. As a result, high-quality networks can be trained in domains with relatively little training data. This property has been extensively studied for discriminative networks but has received statistically-significantly less attention for generative models. Given the often enormous effort required to train GANs, both computationally as well as in the dataset collection, the re-use of pretrained GANs is a desirable objective.

We propose a novel knowledge transfer method for generative models based on mining the knowledge that is most beneficial to a specific target domain, either from a single or multiple pretrained GANs. This is done using a miner network that identifies which part of the generative distribution of each pretrained GAN outputs samples closest to the target domain. Mining effectively steers GAN sampling towards suitable regions of the latent space, which facilitates the posterior finetuning and avoids pathologies of other methods such as mode collapse and lack of flexibility.

We perform experiments on several complex datasets using various GAN architectures (BigGAN⁠, Progressive GAN) and show that the proposed method, called MineGAN, effectively transfers knowledge to domains with few target images, outperforming existing methods. In addition, MineGAN can successfully transfer knowledge from multiple pretrained GANs.

Our code is available at: Github⁠.

Despite an impressive performance from the latest GAN for generating hyper-realistic images, GAN discriminators have difficulty evaluating the quality of an individual generated sample. This is because the task of evaluating the quality of a generated image differs from deciding if an image is real or fake. A generated image could be perfect except in a single area but still be detected as fake. Instead, we propose a novel approach for detecting where errors occur within a generated image. By collaging real images with generated images, we compute for each pixel, whether it belongs to the real distribution or generated distribution. Furthermore, we leverage attention to model long-range dependency; this allows detection of errors which are reasonable locally but not holistically. For evaluation, we show that our error detection can act as a quality metric for an individual image, unlike FID and IS. We leverage Improved Wasserstein, BigGAN⁠, and StyleGAN to show a ranking based on our metric correlates impressively with FID scores. Our work opens the door for better understanding of GAN and the ability to select the best samples from a GAN model.

I generate furry face artwork from color sketches.

The sketches are procedurally generated from a data set of furry artwork. Sketches are translated back into artwork via a Generative Adversarial Network.

I implement the GAN using a U-Net autoencoder with encoder-decoder skip connections and experiment with adding adaptive instance normalization into upsampling layers.

The results show effective mapping of training and dev set sketches back to their input style. However, the model does not perform as effectively on novel user sketches and often fails to add stochastic textures like hair details.

Anime line sketch colorization is to fill a variety of colors the anime sketch, to make it colorful and diverse. The coloring problem is not a new research direction in the field of deep learning technology. Because of coloring of the anime sketch does not have fixed color and we can’t take texture or shadow as reference, so it is difficult to learn and have a certain standard to determine whether it is correct or not. After generative adversarial networks (GANs) was proposed, some used GANs to do coloring research, achieved some result, but the coloring effect is limited.

This study proposes a method use deep residual network, and adding discriminator to network, that expect the color of colored images can consistent with the desired color by the user and can achieve good coloring results.

Generative Adversarial Networks (GANs) are a type of generative model which have received much attention due to their ability to model complex real-world data. Despite their recent successes, the process of training GANs remains challenging, suffering from instability problems such as non-convergence, vanishing or exploding gradients, and mode collapse. In recent years, a diverse set of approaches have been proposed which focus on stabilizing the GAN training procedure. The purpose of this survey is to provide a comprehensive overview of the GAN training stabilization methods which can be found in the literature. We discuss the advantages and disadvantages of each approach, offer a comparative summary, and conclude with a discussion of open problems.

[Artbreeder is an interactive GAN generator website. Originally named “Ganbreeder” and providing only the 256px BigGAN generator, it now provides a variety of BigGAN & StyleGAN models, including the anime portrait StyleGAN model. (It is more general than the similar Waifu Labs, but my anime model is not as good.)

Users can generate random samples and explore slight variants of them to gradually explore the “latent space” and find interesting images, but they can also edit images more directly, upload existing images to find the most similar image produced by the model, etc. A popular website, it has generated >56m images from September 2019 to January 2020.]

We also observe that D’s loss approaches zero during training, but undergoes a sharp upward jump at collapse (Appendix F). One possible explanation for this behavior is that D is overfitting to the training set, memorizing training examples rather than learning some meaningful boundary between real and generated images.

As a simple test for D’s memorization (related to Gulrajani et al 2017), we evaluate uncollapsed discriminators on the ImageNet training and validation sets, and measure what percentage of samples are classified as real or generated. While the training accuracy is consistently above 98%, the validation accuracy falls in the range of 50–55%, no better than random guessing (regardless of regularization strategy). This confirms that D is indeed memorizing the training set; we deem this in line with D’s role, which is not explicitly to generalize, but to distill the training data and provide an useful learning signal for G. Additional experiments and discussion are provided in Appendix G.

Line art colorization is expensive and challenging to automate. A GAN approach is proposed, called Tag2Pix, of line art colorization which takes as input a grayscale line art and color tag information and produces a quality colored image.

First, we present the Tag2Pix line art colorization dataset. A generator network is proposed which consists of convolutional layers to transform the input line art, a pre-trained semantic extraction network, and an encoder for input color information. The discriminator is based on an auxiliary classifier GAN to classify the tag information as well as genuineness. In addition, we propose a novel network structure called SECat, which makes the generator properly colorize even small features such as eyes, and also suggest a novel two-step training method where the generator and discriminator first learn the notion of object and shape and then, based on the learned notion, learn colorization, such as where and how to place which color. We present both quantitative and qualitative evaluations which prove the effectiveness of the proposed method.

Anime sketch coloring is to fill various colors into the black-and-white anime sketches and finally obtain the color anime images. Recently, anime sketch coloring has become a new research hotspot in the field of deep learning. In anime sketch coloring, generative adversarial networks (GANs) have been used to design appropriate coloring methods and achieved some results. However, the existing methods based on GANs generally have low-quality coloring effects, such as unreasonable color mixing, poor color gradient effect. In this paper, an efficient anime sketch coloring method using swish-gated residual U-Net (SGRU) and spectrally normalized GAN (SNGAN) has been proposed to solve the above problems. The proposed method is called spectrally normalized GAN with swish-gated residual U-Net (SSN-GAN). In SSN-GAN, SGRU is used as the generator. SGRU is the U-Net with the proposed swish layer and swish-gated residual blocks (SGBs). In SGRU, the proposed swish layer and swish-gated residual blocks (SGBs) effectively filter the information transmitted by each level and improve the performance of the network. The perceptual loss and the per-pixel loss are used to constitute the final loss of SGRU. The discriminator of SSN-GAN uses spectral normalization as a stabilizer of training of GAN, and it is also used as the perceptual network for calculating the perceptual loss. SSN-GAN can automatically color the sketch without providing any coloring hints in advance and can be easily end-to-end trained. Experimental results show that our method performs better than other state-of-the-art coloring methods, and can obtain colorful anime images with higher visual quality.

The idea that the brain learns generative models of the world has been widely promulgated. Most approaches have assumed that the brain learns an explicit density model that assigns a probability to each possible state of the world. However, explicit density models are difficult to learn, requiring approximate inference techniques that may find poor solutions. An alternative approach is to learn an implicit density model that can sample from the generative model without evaluating the probabilities of those samples. The implicit model can be trained to fool a discriminator into believing that the samples are real. This is the idea behind generative adversarial algorithms, which have proven adept at learning realistic generative models. This paper develops an adversarial framework for probabilistic computation in the brain. It first considers how generative adversarial algorithms overcome some of the problems that vex prior theories based on explicit density models. It then discusses the psychological and neural evidence for this framework, as well as how the breakdown of the generator and discriminator could lead to delusions observed in some mental disorders.

…Our sensory inputs are impoverished, and yet our experience of the world feels richly detailed. For example, our fovea permits us access to a high fidelity region of the visual field only twice the size of our thumbnail held at arm’s length. But we don’t experience the world as though looking through a tiny aperture. Instead, our brains feed us a “grand illusion” of panoptic vision (Chater, 2018; Noe et al 2000; Odegaard et al 2018). Similarly, we receive no visual input in the region of the retina that connects to the optic nerve, yet under normal circumstances we are unaware of this blind spot. Moreover, even when we receive high fidelity visual input, we may still fail to witness dramatic changes in scenes (Simons, 2000), as though our brains have contrived imaginary scenes that displace the true scenes.

…First, how can we explain the phenomenology of illusion: why do some illusions feel real, as though one is actually seeing them, whereas other inferences carry information content without the same perceptual experience. For example, Ramachandran and Hirstein (1997) use the example of gazing at wallpaper in a bathroom, where the wallpaper in your visual periphery is ‘filled in’ (you subjectively experience it as high fidelity even though objectively you perceive it with low fidelity), but the wallpaper behind your head is not filled in. In other words, you infer that the wallpaper continues behind your head, and you may even know this with high confidence, but you do not have the experience of seeing the wallpaper behind your head. Thus, the vividness or “realness” of perceptual experience is not a simple function of belief strength. So what is it a function of? Second, how can we explain the peculiar ways that the inferential apparatus breaks down? In particular, how can we understand the origins of delusions, hallucinations, and confabulations that arise in certain mental disorders? While Bayesian models have been developed to explain these phenomena, they fall short in certain ways that we discuss later on.

An open secret in contemporary machine learning is that many models work beautifully on standard benchmarks but fail to generalize outside the lab. This has been attributed to biased training data, which provide poor coverage over real world events. Generative models are no exception, but recent advances in generative adversarial networks (GANs) suggest otherwise—these models can now synthesize strikingly realistic and diverse images. Is generative modeling of photos a solved problem? We show that although current GANs can fit standard datasets very well, they still fall short of being comprehensive models of the visual manifold. In particular, we study their ability to fit simple transformations such as camera movements and color changes. We find that the models reflect the biases of the datasets on which they are trained (eg. centered objects), but that they also exhibit some capacity for generalization: by “steering” in latent space, we can shift the distribution while still creating realistic images. We hypothesize that the degree of distributional shift is related to the breadth of the training data distribution. Thus, we conduct experiments to quantify the limits of GAN transformations and introduce techniques to mitigate the problem. Code is released on our project page: this URL⁠.

Adversarially trained generative models (GANs) have recently achieved compelling image synthesis results. But despite early successes in using GANs for unsupervised representation learning, they have since been superseded by approaches based on self-supervision. In this work we show that progress in image generation quality translates to substantially improved representation learning performance. Our approach, BigBiGAN, builds upon the state-of-the-art BigGAN model, extending it to representation learning by adding an encoder and modifying the discriminator. We extensively evaluate the representation learning and generation capabilities of these BigBiGAN models, demonstrating that these generation-based models achieve the state of the art in unsupervised representation learning on ImageNet, as well as in unconditional image generation. Pretrained BigBiGAN models—including image generators and encoders—are available on TensorFlow Hub (https:/​/​tfhub.dev/​s?publisher = deepmind&q = bigbigan).

Generative models are models that generate music, images, text, and other complex data types. In recent years generative models have advanced at an astonishing rate, largely due to deep learning, and particularly due to generative adversarial models (GANs). However, GANs are notoriously difficult to train, due to requiring a large amount of data, needing many GPUs and a lot of time to train, and being highly sensitive to minor hyperparameter changes.

fast.ai has been working in recent years towards making a range of models easier and faster to train, with a particular focus on using transfer learning. Transfer learning refers to pre-training a model using readily available data and quick and easy to calculate loss functions, and then fine-tuning that model for a task that may have fewer labels, or be more expensive to compute. This seemed like a potential solution to the GAN training problem, so in late 2018 fast.ai worked on a transfer learning technique for generative modeling⁠.

The pre-trained model that fast.ai selected was this: Start with an image dataset and “crappify” the images, such as reducing the resolution, adding jpeg artifacts, and obscuring parts with random text. Then train a model to “decrappify” those images to return them to their original state. fast.ai started with a model that was pre-trained for ImageNet classification, and added a U-Net upsampling network, adding various modern tweaks to the regular U-Net. A simple fast loss function was initially used: mean squared pixel error. This U-Net could be trained in just a few minutes. Then, the loss function was replaced was a combination of other loss functions used in the generative modeling literature (more details in the f8 video) and trained for another couple of hours. The plan was then to finally add a GAN for the last few epochs—however it turned out that the results were so good that fast.ai ended up not using a GAN for the final models….

NoGAN Training: NoGAN is a new and exciting technique in GAN training that we developed, in pursuit of higher quality and more stable renders. How, and how well, it works is a bit surprising.

Here is the NoGAN training process:

1. Pretrain the Generator. The generator is first trained in a more conventional and easier to control manner—with Perceptual Loss (aka Feature Loss) by itself. GAN training is not introduced yet. At this point you’re training the generator as best as you can in the easiest way possible. This takes up most of the time in NoGAN training. Keep in mind: this pretraining by itself will get the generator model far. Colorization will be well-trained as a task, albeit the colors will tend toward dull tones. Self-Attention will also be well-trained at the at this stage, which is very important.

2. Save Generated Images From Pretrained Generator.

3. Pretrain the Critic as a Binary Classifier. Much like in pretraining the generator, what we aim to achieve in this step is to get as much training as possible for the critic in a more “conventional” manner which is easier to control. And there’s nothing easier than a binary classifier! Here we’re training the critic as a binary classifier of real and fake images, with the fake images being those saved in the previous step. A helpful thing to keep in mind here is that you can simply use a pre-trained critic used for another image-to-image task and refine it. This has already been done for super-resolution, where the critic’s pretrained weights were loaded from that of a critic trained for colorization. All that is needed to make use of the pre-trained critic in this case is a little fine-tuning.

4. Train Generator and Critic in (Almost) Normal GAN Setting. Quickly! This is the surprising part. It turns out that in this pretraining scenario, the critic will rapidly drive adjustments in the generator during GAN training. This happens during a narrow window of time before an “inflection point” of sorts is hit. After this point, there seems to be little to no benefit in training any further in this manner. In fact, if training is continued after this point, you’ll start seeing artifacts and glitches introduced in renderings.

In the case of DeOldify, training to this point requires iterating through only about 1% to 3% of ImageNet data (or roughly 2600 to 7800 iterations on a batch size of five). This amounts to just around 30–90 minutes of GAN training, which is in stark contrast to the three to five days of progressively-sized GAN training that was done previously. Surprisingly, during that short amount of training, the change in the quality of the renderings is dramatic. In fact, this makes up the entirety of GAN training for the video model. The “artistic” and “stable” models go one step further and repeat the NoGAN training process steps 2–4 until there’s no more apparent benefit (around five repeats).

Note: a small but important change to this GAN training that deviates from conventional GANs is the use of a loss threshold that must be met by the critic before generator training commences. Until then, the critic continues training to “catch up” in order to be able to provide the generator with constructive gradients. This catch up chiefly takes place at the beginning of GAN training which immediately follows generator and critic pretraining.

The ability to automatically estimate the quality and coverage of the samples produced by a generative model is a vital requirement for driving algorithm research. We present an evaluation metric that can separately and reliably measure both of these aspects in image generation tasks by forming explicit, non-parametric representations of the manifolds of real and generated data. We demonstrate the effectiveness of our metric in StyleGAN and BigGAN by providing several illustrative examples where existing metrics yield uninformative or contradictory results. Furthermore, we analyze multiple design variants of StyleGAN to better understand the relationships between the model architecture, training methods, and the properties of the resulting sample distribution. In the process, we identify new variants that improve the state-of-the-art. We also perform the first principled analysis of truncation methods and identify an improved method. Finally, we extend our metric to estimate the perceptual quality of individual samples, and use this to study latent space interpolations.

While Generative Adversarial Networks (GANs) have seen huge successes in image synthesis tasks, they are notoriously difficult to adapt to different datasets, in part due to instability during training and sensitivity to hyperparameters. One commonly accepted reason for this instability is that gradients passing from the discriminator to the generator become uninformative when there isn’t enough overlap in the supports of the real and fake distributions. In this work, we propose the Multi-Scale Gradient Generative Adversarial Network (MSG-GAN), a simple but effective technique for addressing this by allowing the flow of gradients from the discriminator to the generator at multiple scales. This technique provides a stable approach for high resolution image synthesis, and serves as an alternative to the commonly used progressive growing technique. We show that MSG-GAN converges stably on a variety of image datasets of different sizes, resolutions and domains, as well as different types of loss functions and architectures, all with the same set of fixed hyperparameters. When compared to state-of-the-art GANs, our approach matches or exceeds the performance in most of the cases we tried.

I describe how I made the website ThisWaifuDoesNotExist.net (TWDNE) for displaying random anime faces generated by StyleGAN neural networks, and how it went viral.

Generating high-quality anime faces has long been a task neural networks struggled with. The invention of StyleGAN in 2018 has effectively solved this task and I have trained a StyleGAN model which can generate high-quality anime faces at 512px resolution. To show off the recent progress, I made a website, “This Waifu Does Not Exist” for displaying random StyleGAN 2 faces. TWDNE displays a different neural-net-generated face & plot summary every 15s. The site was popular and went viral online, especially in China. The model can also be used interactively for exploration & editing in the Artbreeder online service⁠.

TWDNE faces have been used as screensavers, user avatars, character art for game packs or online games⁠, painted watercolors⁠, uploaded to Pixiv, given away in streams⁠, and used in a research paper (Noguchi & Harada 2019). TWDNE results also helped inspired Sizigi Studio’s online interactive waifu GAN⁠, Waifu Labs⁠, which generates even better anime faces than my StyleGAN results.

A tutorial explaining how to train and generate high-quality anime faces with StyleGAN 1/​2 neural networks, and tips/​scripts for effective StyleGAN use.

Generative neural networks, such as GANs, have struggled for years to generate decent-quality anime faces, despite their great success with photographic imagery such as real human faces. The task has now been effectively solved, for anime faces as well as many other domains, by the development of a new generative adversarial network, StyleGAN⁠, whose source code was released in February 2019.

I show off my StyleGAN 1/​2 CC-0-licensed anime faces & videos, provide downloads for the final models & anime portrait face dataset⁠, provide the ‘missing manual’ & explain how I trained them based on Danbooru2017 /  ​ 2018 with source code for the data preprocessing⁠, document installation & configuration & training tricks⁠.

For application, I document various scripts for generating images & videos⁠, briefly describe the website “This Waifu Does Not Exist” I set up as a public demo & its followup This Anime Does Not Exist.ai (TADNE) (see also Artbreeder), discuss how the trained models can be used for transfer learning such as generating high-quality faces of anime characters with small datasets (eg. Holo or Asuka Souryuu Langley), and touch on more advanced StyleGAN applications like encoders & controllable generation.

The anime face graveyard gives samples of my failures with earlier GANs for anime face generation, and I provide samples & model from a relatively large-scale BigGAN training run suggesting that BigGAN may be the next step forward to generating full-scale anime images.

A minute of reading could save an hour of debugging!

Experiments in using BigGAN to generate anime faces and whole anime images; semi-successful.

Following my StyleGAN anime face experiments⁠, I explore BigGAN⁠, another recent GAN with SOTA results on one of the most complex image domains tackled by GANs so far (ImageNet). BigGAN’s capabilities come at a steep compute cost, however.

Using the unofficial BigGAN-PyTorch reimplementation, I experimented in 2019 with 128px ImageNet transfer learning (successful) with ~6 GPU-days, and from-scratch 256px anime portraits of 1000 characters on an 8×2080ti machine for a month (mixed results). My BigGAN results are good but compromised by the compute expense & practical problems with the released BigGAN code base. While BigGAN is not yet superior to StyleGAN for many purposes, BigGAN-like approaches may be necessary to scale to whole anime images.

For followup experiments, Shawn Presser⁠, I and others (collectively, “Tensorfork”) have used Tensorflow Research Cloud TPU credits & the compare_gan BigGAN reimplementation. Running this at scale on the full Danbooru2019 dataset in May 2020, we have reached the best anime GAN results to date (later exceeded by This Anime Does Not Exist).

Generative Adversarial Networks (GANs) have been used in several machine learning tasks such as domain transfer, super resolution, and synthetic data generation. State-of-the-art GANs often use tens of millions of parameters, making them expensive to deploy for applications in low SWAP (size, weight, and power) hardware, such as mobile devices, and for applications with real time capabilities. There has been no work found to reduce the number of parameters used in GANs. Therefore, we propose a method to compress GANs using knowledge distillation techniques, in which a smaller “student” GAN learns to mimic a larger “teacher” GAN. We show that the distillation methods used on MNIST⁠, CIFAR-10, and Celeb-A datasets can compress teacher GANs at ratios of 1669:1, 58:1, and 87:1, respectively, while retaining the quality of the generated image. From our experiments, we observe a qualitative limit for GAN’s compression. Moreover, we observe that, with a fixed parameter budget, compressed GANs outperform GANs trained using standard training methods. We conjecture that this is partially owing to the optimization landscape of over-parameterized GANs which allows efficient training using alternating gradient descent. Thus, training an over-parameterized GAN followed by our proposed compression scheme provides a high quality generative model with a small number of parameters.

Generative Adversarial Networks (GAN) boast impressive capacity to generate realistic images. However, like much of the field of deep learning, they require an inordinate amount of data to produce results, thereby limiting their usefulness in generating novelty. In the same vein, recent advances in meta-learning have opened the door to many few-shot learning applications. In the present work, we propose Few-shot Image Generation using Reptile (FIGR), a GAN meta-trained with Reptile.

Our model successfully generates novel images on both MNIST and Omniglot with as little as 4 images from an unseen class. We further contribute FIGR-8, a new dataset for few-shot image generation, which contains 1,548,944 icons categorized in over 18,409 classes. Trained on FIGR-8, initial results show that our model can generalize to more advanced concepts (such as “bird” and “knife”) from as few as 8 samples from a previously unseen class of images and as little as 10 training steps through those 8 images.

This work demonstrates the potential of training a GAN for few-shot image generation and aims to set a new benchmark for future work in the domain.

We’ve discovered that the gradient noise scale, a simple statistical metric, predicts the parallelizability of neural network training on a wide range of tasks. Since complex tasks tend to have noisier gradients, increasingly large batch sizes are likely to become useful in the future, removing one potential limit to further growth of AI systems. More broadly, these results show that neural network training need not be considered a mysterious art, but can be rigorized and systematized.

In an increasing number of domains it has been demonstrated that deep learning models can be trained using relatively large batch sizes without sacrificing data efficiency. However the limits of this massive data parallelism seem to differ from domain to domain, ranging from batches of tens of thousands in ImageNet to batches of millions in RL agents that play the game Dota 2. To our knowledge there is limited conceptual understanding of why these limits to batch size differ or how we might choose the correct batch size in a new domain. In this paper, we demonstrate that a simple and easy-to-measure statistic called the gradient noise scale predicts the largest useful batch size across many domains and applications, including a number of supervised learning datasets (MNIST, SVHN, CIFAR-10, ImageNet, Billion Word), reinforcement learning domains (Atari and Dota), and even generative model training (autoencoders on SVHN). We find that the noise scale increases as the loss decreases over a training run and depends on the model size primarily through improved model performance. Our empirically-motivated theory also describes the tradeoff between compute-efficiency and time-efficiency, and provides a rough model of the benefits of adaptive batch-size training.

…We have found that by measuring the gradient noise scale, a simple statistic that quantifies the signal-to-noise ratio of the network gradients, we can approximately predict the maximum useful batch size. Heuristically, the noise scale measures the variation in the data as seen by the model (at a given stage in training). When the noise scale is small, looking at a lot of data in parallel quickly becomes redundant, whereas when it is large, we can still learn a lot from huge batches of data…We’ve found it helpful to visualize the results of these experiments in terms of a tradeoff between wall time for training and total bulk compute that we use to do the training (proportional to dollar cost). At very small batch sizes, doubling the batch allows us to train in half the time without using extra compute (we run twice as many chips for half as long). At very large batch sizes, more parallelization doesn’t lead to faster training. There is a “bend” in the curve in the middle, and the gradient noise scale predicts where that bend occurs.

…more powerful models have a higher gradient noise scale, but only because they achieve a lower loss. Thus, there’s some evidence that the increasing noise scale over training isn’t just an artifact of convergence, but occurs because the model gets better. If this is true, then we expect future, more powerful models to have higher noise scale and therefore be more parallelizable. Second, tasks that are subjectively more difficult are also more amenable to parallelization…we have evidence that more difficult tasks and more powerful models on the same task will allow for more radical data-parallelism than we have seen to date, providing a key driver for the continued fast exponential growth in training compute.

We present a novel CNN-based image editing strategy that allows the user to change the semantic information of an image over an arbitrary region by manipulating the feature-space representation of the image in a trained GAN model. We will present two variants of our strategy: (1) spatial conditional batch normalization (sCBN), a type of conditional batch normalization with user-specifiable spatial weight maps, and (2) feature-blending, a method of directly modifying the intermediate features. Our methods can be used to edit both artificial image and real image, and they both can be used together with any GAN with conditional normalization layers. We will demonstrate the power of our method through experiments on various types of GANs trained on different datasets. Code will be available at Github⁠.

We know SGAN may have a risk of gradient vanishing. A significant improvement is WGAN, with the help of 1-Lipschitz constraint on discriminator to prevent from gradient vanishing. Is there any GAN having no gradient vanishing and no 1-Lipschitz constraint on discriminator? We do find one, called GAN-QP.

To construct a new framework of Generative Adversarial Network (GAN) usually includes three steps: 1. choose a probability divergence; 2. convert it into a dual form; 3. play a min-max game. In this articles, we demonstrate that the first step is not necessary. We can analyse the property of divergence and even construct new divergence in dual space directly. As a reward, we obtain a simpler alternative of WGAN: GAN-QP. We demonstrate that GAN-QP have a better performance than WGAN in theory and practice.

Generating high-quality text with sufficient diversity is essential for a wide range of Natural Language Generation (NLG) tasks. Maximum-Likelihood (MLE) models trained with teacher forcing have consistently been reported as weak baselines, where poor performance is attributed to exposure bias (Bengio et al 2015; Ranzato et al 2015); at inference time, the model is fed its own prediction instead of a ground-truth token, which can lead to accumulating errors and poor samples. This line of reasoning has led to an outbreak of adversarial based approaches for NLG, on the account that GANs do not suffer from exposure bias. In this work, we make several surprising observations which contradict common beliefs. First, we revisit the canonical evaluation framework for NLG, and point out fundamental flaws with quality-only evaluation: we show that one can outperform such metrics using a simple, well-known temperature parameter to artificially reduce the entropy of the model’s conditional distributions. Second, we leverage the control over the quality / diversity trade-off given by this parameter to evaluate models over the whole quality-diversity spectrum and find MLE models constantly outperform the proposed GAN variants over the whole quality-diversity space. Our results have several implications: (1) The impact of exposure bias on sample quality is less severe than previously thought, (2) temperature tuning provides a better quality / diversity trade-off than adversarial training while being easier to train, easier to cross-validate, and less computationally expensive. Code to reproduce the experiments is available at github.com/​pclucas14/​GansFallingShort

We propose a rejection sampling scheme using the discriminator of a GAN to correct errors in the GAN generator distribution. We show that under quite strict assumptions, this will allow us to recover the data distribution exactly. We then examine where those strict assumptions break down and design a practical algorithm—called Discriminator Rejection Sampling (DRS)—that can be used on real data-sets.

Finally, we demonstrate the efficacy of DRS on a mixture of Gaussians and on the SAGAN model, state-of-the-art in the image generation task at the time of developing this work. On ImageNet, we train an improved baseline that increases the Inception Score from 52.52 to 62.36 and reduces the Fréchet Inception Distance from 18.65 to 14.79. We then use DRS to further improve on this baseline, improving the Inception Score to 76.08 and the FID to 13.75.

Training Generative Adversarial Networks (GANs) is notoriously challenging.

We propose and study an architectural modification, self-modulation, which improves GAN performance across different data sets, architectures, losses, regularizers, and hyperparameter settings. Intuitively, self-modulation allows the intermediate feature maps of a generator to change as a function of the input noise vector. While reminiscent of other conditioning techniques, it requires no labeled data.

In a large-scale empirical study we observe a relative decrease of 5%–35% in FID. Furthermore, all else being equal, adding this modification to the generator leads to improved performance in 124⁄144 (86%) of the studied settings.

Self-modulation is a simple architectural change that requires no additional parameter tuning, which suggests that it can be applied readily to any GAN.

Adversarial learning methods have been proposed for a wide range of applications, but the training of adversarial models can be notoriously unstable. Effectively balancing the performance of the generator and discriminator is critical, since a discriminator that achieves very high accuracy will produce relatively uninformative gradients. In this work, we propose a simple and general technique to constrain information flow in the discriminator by means of an information bottleneck. By enforcing a constraint on the mutual information between the observations and the discriminator’s internal representation, we can effectively modulate the discriminator’s accuracy and maintain useful and informative gradients.

We demonstrate that our proposed variational discriminator bottleneck (VDB) leads to significant improvements across three distinct application areas for adversarial learning algorithms. Our primary evaluation studies the applicability of the VDB to imitation learning of dynamic continuous control skills, such as running. We show that our method can learn such skills directly from raw video demonstrations, substantially outperforming prior adversarial imitation learning methods. The VDB can also be combined with adversarial inverse reinforcement learning to learn parsimonious reward functions that can be transferred and re-optimized in new settings. Finally, we demonstrate that VDB can train GANs more effectively for image generation, improving upon a number of prior stabilization methods.

…To confirm that our design choices are effective for even larger and more complex and diverse datasets, we also present results of our system on a subset of JFT-300M (Sun et al 2017). The full JFT-300M dataset contains 300M real-world images labeled with 18K categories. Since the category distribution is heavily long-tailed, we subsample the dataset to keep only images with the 8.5K most common labels. The resulting dataset contains 292M images—two orders of magnitude larger than ImageNet.

…Our results show that these techniques substantially improve performance even in the setting of this much larger dataset at the same model capacity (64 base channels). We further show that for a dataset of this scale, we see substantial additional improvements from expanding the capacity of our models to 128 base channels, while for ImageNet GANs that additional capacity was not beneficial. In Figure 19 (Appendix D), we present truncation plots for models trained on this dataset…Interestingly, unlike models trained on ImageNet, where training tends to collapse without heavy regularization (Section 4), the models trained on JFT-300M remain stable over many hundreds of thousands of iterations. This suggests that moving beyond ImageNet to larger datasets may partially alleviate GAN stability issues.

Despite recent progress in generative image modeling, successfully generating high-resolution, diverse samples from complex datasets such as ImageNet remains an elusive goal. To this end, we train Generative Adversarial Networks at the largest scale yet attempted, and study the instabilities specific to such scale. We find that applying orthogonal regularization to the generator renders it amenable to a simple “truncation trick”, allowing fine control over the trade-off between sample fidelity and variety by reducing the variance of the Generator’s input. Our modifications lead to models which set the new state of the art in class-conditional image synthesis. When trained on ImageNet at 128×128 resolution, our models (BigGANs) achieve an Inception Score (IS) of 166.5 and Fréchet Inception Distance (FID) of 7.4, improving over the previous best IS of 52.52 and FID of 18.6.

The Super-Resolution Generative Adversarial Network (SRGAN) is a seminal work that is capable of generating realistic textures during single image super-resolution. However, the hallucinated details are often accompanied with unpleasant artifacts.

To further enhance the visual quality, we thoroughly study three key components of SRGAN—network architecture, adversarial loss and perceptual loss, and improve each of them to derive an Enhanced SRGAN (ESRGAN). In particular, we introduce the Residual-in-Residual Dense Block (RRDB) without batch normalization as the basic network building unit. Moreover, we borrow the idea from relativistic GAN to let the discriminator predict relative realness instead of the absolute value. Finally, we improve the perceptual loss by using the features before activation, which could provide stronger supervision for brightness consistency and texture recovery.

Benefiting from these improvements, the proposed ESRGAN achieves consistently better visual quality with more realistic and natural textures than SRGAN and won the first place in the PIRM2018-SR Challenge.

The code is available at Github⁠.

We present a framework for translating unlabeled images from one domain into analog images in another domain. We employ a progressively growing skip-connected encoder-generator structure and train it with a GAN loss for realistic output, a cycle consistency loss for maintaining same-domain translation identity, and a semantic consistency loss that encourages the network to keep the input semantic features in the output. We apply our framework on the task of translating face images, and show that it is capable of learning semantic mappings for face images with no supervised one-to-one image mapping.

We present a novel introspective variational autoencoder (IntroVAE) model for synthesizing high-resolution photographic images. IntroVAE is capable of self-evaluating the quality of its generated samples and improving itself accordingly. Its inference and generator models are jointly trained in an introspective way. On one hand, the generator is required to reconstruct the input images from the noisy outputs of the inference model as normal VAEs. On the other hand, the inference model is encouraged to classify between the generated and real samples while the generator tries to fool it as GANs. These two famous generative frameworks are integrated in a simple yet efficient single-stream architecture that can be trained in a single stage. IntroVAE preserves the advantages of VAEs, such as stable training and nice latent manifold. Unlike most other hybrid models of VAEs and GANs, IntroVAE requires no extra discriminators, because the inference model itself serves as a discriminator to distinguish between the generated and real samples. Experiments demonstrate that our method produces high-resolution photo-realistic images (eg. CELEBA images at 1024²), which are comparable to or better than the state-of-the-art GANs.

Unpaired image-to-image translation is the problem of mapping an image in the source domain to one in the target domain, without requiring corresponding image pairs. To ensure the translated images are realistically plausible, recent works, such as Cycle-GAN, demands this mapping to be invertible. While, this requirement demonstrates promising results when the domains are unimodal, its performance is unpredictable in a multi-modal scenario such as in an image segmentation task. This is because, invertibility does not necessarily enforce semantic correctness. To this end, we present a semantically-consistent GAN framework, dubbed Sem-GAN, in which the semantics are defined by the class identities of image segments in the source domain as produced by a semantic segmentation algorithm. Our proposed framework includes consistency constraints on the translation task that, together with the GAN loss and the cycle-constraints, enforces that the images when translated will inherit the appearances of the target domain, while (approximately) maintaining their identities from the source domain. We present experiments on several image-to-image translation tasks and demonstrate that Sem-GAN improves the quality of the translated images significantly, sometimes by more than 20% on the FCN score. Further, we show that semantic segmentation models, trained with synthetic images translated via Sem-GAN, leads to significantly better segmentation results than other variants.

Cartoon Set is a collection of random, 2D cartoon avatar images. The cartoons vary in 10 artwork categories, 4 color categories, and 4 proportion categories, with a total of ~10¹³ possible combinations. We provide sets of 10k and 100k randomly chosen cartoons and labeled attributes…We’ve also used the dataset to research cross-domain image translation: “XGAN: Unsupervised Image-to-Image Translation for Many-to-Many Mappings”⁠.

In standard generative adversarial network (SGAN), the discriminator estimates the probability that the input data is real. The generator is trained to increase the probability that fake data is real. We argue that it should also simultaneously decrease the probability that real data is real because (1) this would account for a priori knowledge that half of the data in the mini-batch is fake, (2) this would be observed with divergence minimization, and (3) in optimal settings, SGAN would be equivalent to integral probability metric (IPM) GANs.

We show that this property can be induced by using a relativistic discriminator which estimate the probability that the given real data is more realistic than a randomly sampled fake data. We also present a variant in which the discriminator estimate the probability that the given real data is more realistic than fake data, on average. We generalize both approaches to non-standard GAN loss functions and we refer to them respectively as Relativistic GANs (RGANs) and Relativistic average GANs (RaGANs). We show that IPM-based GANs are a subset of RGANs which use the identity function.

Empirically, we observe that (1) RGANs and RaGANs are statistically-significantly more stable and generate higher quality data samples than their non-relativistic counterparts, (2) Standard RaGAN with gradient penalty generate data of better quality than WGAN-GP while only requiring a single discriminator update per generator update (reducing the time taken for reaching the state-of-the-art by 400%), and (3) RaGANs are able to generate plausible high resolutions images (256×256) from a very small sample (n = 2011), while GAN and LSGAN cannot; these images are of significantly better quality than the ones generated by WGAN-GP and SGAN with spectral normalization.

We examine two different techniques for parameter averaging in GAN training. Moving Average (MA) computes the time-average of parameters, whereas Exponential Moving Average (EMA) computes an exponentially discounted sum. Whilst MA is known to lead to convergence in bilinear settings, we provide the—to our knowledge—first theoretical arguments in support of EMA. We show that EMA converges to limit cycles around the equilibrium with vanishing amplitude as the discount parameter approaches one for simple bilinear games and also enhances the stability of general GAN training. We establish experimentally that both techniques are strikingly effective in the non-convex-concave GAN setting as well. Both improve inception and FID scores on different architectures and for different GAN objectives. We provide comprehensive experimental results across a range of datasets—mixture of Gaussians, CIFAR-10, STL-10, CelebA and ImageNet—to demonstrate its effectiveness. We achieve state-of-the-art results on CIFAR-10 and produce clean CelebA face images. The code is available at https://github.com/yasinyazici/EMA_GAN⁠.

In this paper, we propose the Self-Attention Generative Adversarial Network (SAGAN) which allows attention-driven, long-range dependency modeling for image generation tasks. Traditional convolutional GANs generate high-resolution details as a function of only spatially local points in lower-resolution feature maps. In SAGAN, details can be generated using cues from all feature locations. Moreover, the discriminator can check that highly detailed features in distant portions of the image are consistent with each other. Furthermore, recent work has shown that generator conditioning affects GAN performance. Leveraging this insight, we apply spectral normalization to the GAN generator and find that this improves training dynamics. The proposed SAGAN achieves the state-of-the-art results, boosting the best published Inception score from 36.8 to 52.52 and reducing Fréchet Inception distance from 27.62 to 18.65 on the challenging ImageNet dataset. Visualization of the attention layers shows that the generator leverages neighborhoods that correspond to object shapes rather than local regions of fixed shape.

Advances in deep generative networks have led to impressive results in recent years. Nevertheless, such models can often waste their capacity on the minutiae of datasets, presumably due to weak inductive biases in their decoders. This is where graphics engines may come in handy since they abstract away low-level details and represent images as high-level programs. Current methods that combine deep learning and renderers are limited by hand-crafted likelihood or distance functions, a need for large amounts of supervision, or difficulties in scaling their inference algorithms to richer datasets. To mitigate these issues, we present SPIRAL, an adversarially trained agent that generates a program which is executed by a graphics engine to interpret and sample images. The goal of this agent is to fool a discriminator network that distinguishes between real and rendered data, trained with a distributed reinforcement learning setup without any supervision. A surprising finding is that using the discriminator’s output as a reward signal is the key to allow the agent to make meaningful progress at matching the desired output rendering. To the best of our knowledge, this is the first demonstration of an end-to-end, unsupervised and adversarial inverse graphics agent on challenging real world (MNIST, OMNIGLOT, CELEBA) and synthetic 3D datasets. A video of the agent can be found at YouTube⁠.

Generative adversarial networks (GANs) form a generative modeling approach known for producing appealing samples, but they are notably difficult to train. One common way to tackle this issue has been to propose new formulations of the GAN objective. Yet, surprisingly few studies have looked at optimization methods designed for this adversarial training. In this work, we cast GAN optimization problems in the general variational inequality framework. Tapping into the mathematical programming literature, we counter some common misconceptions about the difficulties of saddle point optimization and propose to extend techniques designed for variational inequalities to the training of GANs. We apply averaging, extrapolation and a computationally cheaper variant that we call extrapolation from the past to the stochastic gradient method (SGD) and Adam.

One of the challenges in the study of generative adversarial networks is the instability of its training. In this paper, we propose a novel weight normalization technique called spectral normalization to stabilize the training of the discriminator. Our new normalization technique is computationally light and easy to incorporate into existing implementations. We tested the efficacy of spectral normalization on CIFAR10, STL-10, and ILSVRC2012 dataset, and we experimentally confirmed that spectrally normalized GANs (SN-GANs) is capable of generating images of better or equal quality relative to the previous training stabilization techniques.

Generative adversarial networks (GANs) learn a deep generative model that is able to synthesise novel, high-dimensional data samples. New data samples are synthesised by passing latent samples, drawn from a chosen prior distribution, through the generative model. Once trained, the latent space exhibits interesting properties, that may be useful for down stream tasks such as classification or retrieval. Unfortunately, GANs do not offer an “inverse model”, a mapping from data space back to latent space, making it difficult to infer a latent representation for a given data sample. In this paper, we introduce a technique, inversion, to project data samples, specifically images, to the latent space using a pre-trained GAN. Using our proposed inversion technique, we are able to identify which attributes of a dataset a trained GAN is able to model and quantify GAN performance, based on a reconstruction loss. We demonstrate how our proposed inversion technique may be used to quantitatively compare performance of various GAN models trained on three image datasets. We provide code for all of our experiments, Github⁠.

This work details CipherGAN, an architecture inspired by CycleGAN used for inferring the underlying cipher mapping given banks of unpaired ciphertext and plaintext. We demonstrate that CipherGAN is capable of cracking language data enciphered using shift and Vigenere ciphers to a high degree of fidelity and for vocabularies much larger than previously achieved. We present how CycleGAN can be made compatible with discrete data and train in a stable way. We then prove that the technique used in CipherGAN avoids the common problem of uninformative discrimination associated with GANs applied to discrete data.

CycleGAN (Zhu et al 2017) is one recent successful approach to learn a transformation between two image distributions. In a series of experiments, we demonstrate an intriguing property of the model: CycleGAN learns to “hide” information about a source image into the images it generates in a nearly imperceptible, high-frequency signal. This trick ensures that the generator can recover the original sample and thus satisfy the cyclic consistency requirement, while the generated image remains realistic. We connect this phenomenon with adversarial attacks by viewing CycleGAN’s training procedure as training a generator of adversarial examples and demonstrate that the cyclic consistency loss causes CycleGAN to be especially vulnerable to adversarial attacks.

In this work, we focus on the challenge of taking partial observations of highly-stylized text and generalizing the observations to generate unobserved glyphs in the ornamented typeface. To generate a set of multi-content images following a consistent style from very few examples, we propose an end-to-end stacked conditional GAN model considering content along channels and style along network layers. Our proposed network transfers the style of given glyphs to the contents of unseen ones, capturing highly stylized fonts found in the real-world such as those on movie posters or infographics. We seek to transfer both the typographic stylization (ex. serifs and ears) as well as the textual stylization (ex. color gradients and effects.) We base our experiments on our collected data set including 10,000 fonts with different styles and demonstrate effective generalization from a very small number of observed glyphs.

We present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). Conditional GANs have enabled a variety of applications, but the results are often limited to low-resolution and still far from realistic. In this work, we generate 2048×1024 visually appealing results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures. Furthermore, we extend our framework to interactive visual manipulation with two additional features. First, we incorporate object instance segmentation information, which enables object manipulations such as removing/​adding objects and changing the object category. Second, we propose a method to generate diverse results given the same input, allowing users to edit the object appearance interactively. Human opinion studies demonstrate that our method significantly outperforms existing methods, advancing both the quality and the resolution of deep image synthesis and editing.

Generative adversarial networks (GAN) are a powerful subclass of generative models. Despite a very rich research activity leading to numerous interesting GAN algorithms, it is still very hard to assess which algorithm(s) perform better than others. We conduct a neutral, multi-faceted large-scale empirical study on state-of-the art models and evaluation measures. We find that most models can reach similar scores with enough hyperparameter optimization and random restarts. This suggests that improvements can arise from a higher computational budget and tuning more than fundamental algorithmic changes. To overcome some limitations of the current metrics, we also propose several data sets on which precision and recall can be computed. Our experimental results suggest that future GAN research should be based on more systematic and objective evaluation procedures. Finally, we did not find evidence that any of the tested algorithms consistently outperforms the non-saturating GAN introduced in Goodfellow et al 2014.

Style transfer is an important problem in natural language processing (NLP). However, the progress in language style transfer is lagged behind other domains, such as computer vision, mainly because of the lack of parallel data and principle evaluation metrics. In this paper, we propose to learn style transfer with non-parallel data. We explore two models to achieve this goal, and the key idea behind the proposed models is to learn separate content representations and style representations using adversarial networks. We also propose novel evaluation metrics which measure two aspects of style transfer: transfer strength and content preservation. We access our models and the evaluation metrics on two tasks: paper-news title transfer, and positive-negative review transfer. Results show that the proposed content preservation metric is highly correlate to human judgments, and the proposed models are able to generate sentences with higher style transfer strength and similar content preservation score comparing to auto-encoder.

Style transfer usually refers to the task of applying color and texture information from a specific style image to a given content image while preserving the structure of the latter. Here we tackle the more generic problem of semantic style transfer: given two unpaired collections of images, we aim to learn a mapping between the corpus-level style of each collection, while preserving semantic content shared across the two domains. We introduce XGAN (“Cross-GAN”), a dual adversarial autoencoder, which captures a shared representation of the common domain semantic content in an unsupervised way, while jointly learning the domain-to-domain image translations in both directions. We exploit ideas from the domain adaptation literature and define a semantic consistency loss which encourages the model to preserve semantics in the learned embedding space. We report promising qualitative results for the task of face-to-cartoon translation. The cartoon dataset, CartoonSet, we collected for this purpose is publicly available at google.github.io/​cartoonset/ as a new benchmark for semantic style transfer.

We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively: starting from a low resolution, we add new layers that model increasingly fine details as training progresses. This both speeds the training up and greatly stabilizes it, allowing us to produce images of unprecedented quality, eg. CelebA images at 1024^2. We also propose a simple way to increase the variation in generated images, and achieve a record inception score of 8.80 in unsupervised CIFAR10. Additionally, we describe several implementation details that are important for discouraging unhealthy competition between the generator and discriminator. Finally, we suggest a new metric for evaluating GAN results, both in terms of image quality and variation. As an additional contribution, we construct a higher-quality version of the CelebA dataset.

Although Generative Adversarial Networks (GANs) have shown remarkable success in various tasks, they still face challenges in generating high quality images. In this paper, we propose Stacked Generative Adversarial Networks (StackGAN) aiming at generating high-resolution photo-realistic images. First, we propose a two-stage generative adversarial network architecture, StackGAN-v1, for text-to-image synthesis. The Stage-I GAN sketches the primitive shape and colors of the object based on given text description, yielding low-resolution images. The Stage-II GAN takes Stage-I results and text descriptions as inputs, and generates high-resolution images with photo-realistic details. Second, an advanced multi-stage generative adversarial network architecture, StackGAN-v2, is proposed for both conditional and unconditional generative tasks. Our StackGAN-v2 consists of multiple generators and discriminators in a tree-like structure; images at multiple scales corresponding to the same scene are generated from different branches of the tree. StackGAN-v2 shows more stable training behavior than StackGAN-v1 by jointly approximating multiple distributions. Extensive experiments demonstrate that the proposed stacked generative adversarial networks significantly outperform other state-of-the-art methods in generating photo-realistic images.

Instrumenting and collecting annotated visual grasping datasets to train modern machine learning algorithms can be extremely time-consuming and expensive. An appealing alternative is to use off-the-shelf simulators to render synthetic data for which ground-truth annotations are generated automatically. Unfortunately, models trained purely on simulated data often fail to generalize to the real world. We study how randomized simulated environments and domain adaptation methods can be extended to train a grasping system to grasp novel objects from raw monocular RGB images. We extensively evaluate our approaches with a total of more than 25,000 physical test grasps, studying a range of simulation conditions and domain adaptation methods, including a novel extension of pixel-level domain adaptation that we term the GraspGAN. We show that, by using synthetic data and domain adaptation, we are able to reduce the number of real-world samples needed to achieve a given level of performance by up to 50×, using only randomly generated simulated objects. We also show that by using only unlabeled real-world data and our GraspGAN methodology, we obtain real-world grasping performance without any real-world labels that is similar to that achieved with 939,777 labeled real-world samples.

Reinforcement learning has shown promise in learning policies that can solve complex problems. However, manually specifying a good reward function can be difficult, especially for intricate tasks. Inverse reinforcement learning offers an useful paradigm to learn the underlying reward function directly from expert demonstrations. Yet in reality, the corpus of demonstrations may contain trajectories arising from a diverse set of underlying reward functions rather than a single one. Thus, in inverse reinforcement learning, it is useful to consider such a decomposition. The options framework in reinforcement learning is specifically designed to decompose policies in a similar light. We therefore extend the options framework and propose a method to simultaneously recover reward options in addition to policy options. We leverage adversarial methods to learn joint reward-policy options using only observed expert states. We show that this approach works well in both simple and complex continuous control tasks and shows significant performance increases in one-shot transfer learning.

State-of-the-art password guessing tools, such as Hashcat and John the Ripper, enable users to check billions of passwords per second against password hashes. In addition to performing straightforward dictionary attacks, these tools can expand password dictionaries using password generation rules, such as concatenation of words (eg. “password123456”) and leet speak (eg. “password” becomes “p4s5w0rd”). Although these rules work well in practice, expanding them to model further passwords is a laborious task that requires specialized expertise. To address this issue, in this paper we introduce PassGAN, a novel approach that replaces human-generated password rules with theory-grounded machine learning algorithms. Instead of relying on manual password analysis, PassGAN uses a Generative Adversarial Network (GAN) to autonomously learn the distribution of real passwords from actual password leaks, and to generate high-quality password guesses. Our experiments show that this approach is very promising. When we evaluated PassGAN on two large password datasets, we were able to surpass rule-based and state-of-the-art machine learning password guessing tools. However, in contrast with the other tools, PassGAN achieved this result without any a-priori knowledge on passwords or common password structures. Additionally, when we combined the output of PassGAN with the output of Hashcat, we were able to match 51%–73% more passwords than with Hashcat alone. This is remarkable, because it shows that PassGAN can autonomously extract a considerable number of password properties that current state-of-the art rules do not encode.

Automatic generation of facial images has been well studied after the Generative Adversarial Network (GAN) came out. There exists some attempts applying the GAN model to the problem of generating facial images of anime characters, but none of the existing work gives a promising result. In this work, we explore the training of GAN models specialized on an anime facial image dataset. We address the issue from both the data and the model aspect, by collecting a more clean, well-suited dataset and leverage proper, empirical application of DRAGAN. With quantitative analysis and case studies we demonstrate that our efforts lead to a stable and high-quality model. Moreover, to assist people with anime character design, we build a website (http:/​/​make.girls.moe) with our pre-trained model available online, which makes the model easily accessible to general public.

Neural networks are known to be vulnerable to adversarial examples, inputs that have been intentionally perturbed to remain visually similar to the source input, but cause a misclassification. It was recently shown that given a dataset and classifier, there exists so called universal adversarial perturbations, a single perturbation that causes a misclassification when applied to any input. In this work, we introduce universal adversarial networks, a generative network that is capable of fooling a target classifier when it’s generated output is added to a clean sample from a dataset. We show that this technique improves on known universal adversarial attacks.

We propose a technique for multi-task learning from demonstration that trains the controller of a low-cost robotic arm to accomplish several complex picking and placing tasks, as well as non-prehensile manipulation. The controller is a recurrent neural network using raw images as input and generating robot arm trajectories, with the parameters shared across the tasks. The controller also combines VAE-GAN-based reconstruction with autoregressive multimodal action prediction. Our results demonstrate that it is possible to learn complex manipulation tasks, such as picking up a towel, wiping an object, and depositing the towel to its previous position, entirely from raw images with direct behavior cloning. We show that weight sharing and reconstruction-based regularization substantially improve generalization and robustness, and training on multiple tasks simultaneously increases the success rate on all tasks.

Material recognition enables robots to incorporate knowledge of material properties into their interactions with everyday objects. For example, material recognition opens up opportunities for clearer communication with a robot, such as “bring me the metal coffee mug”, and recognizing plastic versus metal is crucial when using a microwave or oven. However, collecting labeled training data with a robot is often more difficult than unlabeled data. We present a semi-supervised learning approach for material recognition that uses generative adversarial networks (GANs) with haptic features such as force, temperature, and vibration. Our approach achieves state-of-the-art results and enables a robot to estimate the material class of household objects with ~90% accuracy when 92% of the training data are unlabeled. We explore how well this approach can recognize the material of new objects and we discuss challenges facing generalization. To motivate learning from unlabeled training data, we also compare results against several common supervised learning classifiers. In addition, we have released the dataset used for this work which consists of time-series haptic measurements from a robot that conducted thousands of interactions with 72 household objects.

We propose a new system for generating art. The system generates art by looking at art and learning about style; and becomes creative by increasing the arousal potential of the generated art by deviating from the learned styles. We build over Generative Adversarial Networks (GAN), which have shown the ability to learn to generate novel images simulating a given distribution. We argue that such networks are limited in their ability to generate creative products in their original design. We propose modifications to its objective to make it capable of generating creative art by maximizing deviation from established styles and minimizing deviation from art distribution.

We conducted experiments to compare the response of human subjects to the generated art with their response to art created by artists. The results show that human subjects could not distinguish art generated by the proposed system from art generated by contemporary artists and shown in top art fairs. Human subjects even rated the generated images higher on various scales.

Generative Adversarial Networks (GANs) have shown great promise recently in image generation. Training GANs for language generation has proven to be more difficult, because of the non-differentiable nature of generating text with recurrent neural networks. Consequently, past work has either resorted to pre-training with maximum-likelihood or used convolutional networks for generation. In this work, we show that recurrent neural networks can be trained to generate text with GANs from scratch using curriculum learning, by slowly teaching the model to generate sequences of increasing and variable length. We empirically show that our approach vastly improves the quality of generated sequences compared to a convolutional baseline.

Since its appearance, Generative Adversarial Networks (GANs) have received a lot of interest in the AI community. In image generation several projects showed how GANs are able to generate photorealistic images but the results so far did not look adequate for the quality standard of visual media production industry. We present an optimized image generation process based on a Deep Convolutional Generative Adversarial Networks (DCGANs), in order to create photorealistic high-resolution images (up to 1024×1024 pixels). Furthermore, the system was fed with a limited dataset of images, less than two thousand images. All these results give more clue about future exploitation of GANs in Computer Graphics and Visual Effects.

We investigate the generalizability of deep learning based on the sensitivity to input perturbation. We hypothesize that the high sensitivity to the perturbation of data degrades the performance on it. To reduce the sensitivity to perturbation, we propose a simple and effective regularization method, referred to as spectral norm regularization, which penalizes the high spectral norm of weight matrices in neural networks. We provide supportive evidence for the above-mentioned hypothesis by experimentally confirming that the models trained using spectral norm regularization exhibit better generalizability than other baseline methods.

In unsupervised data generation tasks, besides the generation of a sample based on previous observations, one would often like to give hints to the model in order to bias the generation towards desirable metrics. We propose a method that combines Generative Adversarial Networks (GANs) and reinforcement learning (RL) in order to accomplish exactly that. While RL biases the data generation process towards arbitrary metrics, the GAN component of the reward function ensures that the model still remembers information learned from data. We build upon previous results that incorporated GANs and RL in order to generate sequence data and test this model in several settings for the generation of molecules encoded as text sequences (SMILES) and in the context of music generation, showing for each case that we can effectively bias the generation process towards desired metrics.

Deep generative models based on Generative Adversarial Networks (GANs) have demonstrated impressive sample quality but in order to work they require a careful choice of architecture, parameter initialization, and selection of hyper-parameters. This fragility is in part due to a dimensional mismatch or non-overlapping support between the model distribution and the data distribution, causing their density ratio and the associated f-divergence to be undefined. We overcome this fundamental limitation and propose a new regularization approach with low computational cost that yields a stable GAN training procedure. We demonstrate the effectiveness of this regularizer across several architectures trained on common benchmark image generation tasks. Our regularization turns GAN models into reliable building blocks for deep learning.

Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to undesired behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.

Translating information between text and image is a fundamental problem in artificial intelligence that connects natural language processing and computer vision.

In the past few years, performance in image caption generation has seen substantial improvement through the adoption of recurrent neural networks (RNN). Meanwhile, text-to-image generation begun to generate plausible images using datasets of specific categories like birds and flowers. We’ve even seen image generation from multi-category datasets such as the Microsoft Common Objects in Context (MS-COCO) through the use of generative adversarial networks (GANs). Synthesizing objects with a complex shape, however, is still challenging. For example, animals and humans have many degrees of freedom, which means that they can take on many complex shapes.

We propose a new training method called Image-Text-Image (I2T2I) which integrates text-to-image and image-to-text (image captioning) synthesis to improve the performance of text-to-image synthesis. We demonstrate that I2T2I can generate better multi-categories using MS COCO than the state-of-the-art. We also demonstrate that I2T2I can achieve transfer learning by using a pre-trained image captioning module to generate human images on the MPII Human Pose dataset (MHP) without using sentence annotations.

While humans easily recognize relations between data from different domains without any supervision, learning to automatically discover them is in general very challenging and needs many ground-truth pairs that illustrate the relations. To avoid costly pairing, we address the task of discovering cross-domain relations given unpaired data. We propose a method based on generative adversarial networks that learns to discover relations between different domains (DiscoGAN). Using the discovered relations, our proposed network successfully transfers style from one domain to another while preserving key attributes such as orientation and face identity. Source code for official implementation is publicly available https://github.com/SKTBrain/DiscoGAN

This paper proposes an approach for applying GANs to NMT. We build a conditional sequence generative adversarial net which comprises of two adversarial sub models, a generator and a discriminator. The generator aims to generate sentences which are hard to be discriminated from human-translated sentences (ie. the golden target sentences), And the discriminator makes efforts to discriminate the machine-generated sentences from human-translated ones. The two sub models play a mini-max game and achieve the win-win situation when they reach a Nash Equilibrium. Additionally, the static sentence-level BLEU is utilized as the reinforced objective for the generator, which biases the generation towards high BLEU points. During training, both the dynamic discriminator and the static BLEU objective are employed to evaluate the generated sentences and feedback the evaluations to guide the learning of the generator. Experimental results show that the proposed model consistently outperforms the traditional RNNSearch and the newly emerged state-of-the-art Transformer on English-German and Chinese-English translation tasks.

This paper proposes an extension to the Generative Adversarial Networks (GANs), namely as ARTGAN to synthetically generate more challenging and complex images such as artwork that have abstract characteristics. This is in contrast to most of the current solutions that focused on generating natural images such as room interiors, birds, flowers and faces. The key innovation of our work is to allow back-propagation of the loss function w.r.t. the labels (randomly assigned to each generated images) to the generator from the discriminator. With the feedback from the label information, the generator is able to learn faster and achieve better generated image quality. Empirically, we show that the proposed ARTGAN is capable to create realistic artwork, as well as generate compelling real world images that globally look natural with clear shape on CIFAR-10.

We introduce a new algorithm named WGAN, an alternative to traditional GAN training. In this new model, we show that we can improve the stability of learning, get rid of problems like mode collapse, and provide meaningful learning curves useful for debugging and hyperparameter searches. Furthermore, we show that the corresponding optimization problem is sound, and provide extensive theoretical work highlighting the deep connections to other distances between distributions.

This report summarizes the tutorial presented by the author at NIPS 2016 on generative adversarial networks (GANs). The tutorial describes: (1) Why generative modeling is a topic worth studying, (2) how generative models work, and how GANs compare to other generative models, (3) the details of how GANs work, (4) research frontiers in GANs, and (5) state-of-the-art image models that combine GANs with other methods. Finally, the tutorial contains three exercises for readers to complete, and the solutions to these exercises.

In this paper, we propose a novel generative model named Stacked Generative Adversarial Networks (SGAN), which is trained to invert the hierarchical representations of a bottom-up discriminative network. Our model consists of a top-down stack of GANs, each learned to generate lower-level representations conditioned on higher-level representations. A representation discriminator is introduced at each feature hierarchy to encourage the representation manifold of the generator to align with that of the bottom-up discriminative network, leveraging the powerful discriminative representations to guide the generative model. In addition, we introduce a conditional loss that encourages the use of conditional information from the layer above, and a novel entropy loss that maximizes a variational lower bound on the conditional entropy of generator outputs.

We first train each stack independently, and then train the whole model end-to-end. Unlike the original GAN that uses a single noise vector to represent all the variations, our SGAN decomposes variations into multiple levels and gradually resolves uncertainties in the top-down generative process. Based on visual inspection, Inception scores and visual Turing test, we demonstrate that SGAN is able to generate images of much higher quality than GANs without stacking.

Generative Adversarial Networks have become one of the most studied frameworks for unsupervised learning due to their intuitive formulation. They have also been shown to be capable of generating convincing examples in limited domains, such as low-resolution images. However, they still prove difficult to train in practice and tend to ignore modes of the data generating distribution. Quantitatively capturing effects such as mode coverage and more generally the quality of the generative model still remain elusive. We propose Generative Adversarial Parallelization, a framework in which many GANs or their variants are trained simultaneously, exchanging their discriminators. This eliminates the tight coupling between a generator and discriminator, leading to improved convergence and improved coverage of modes. We also propose an improved variant of the recently proposed Generative Adversarial Metric and show how it can score individual GANs or their collections under the GAP model.

Synthesizing high-quality images from text descriptions is a challenging problem in computer vision and has many practical applications. Samples generated by existing text-to-image approaches can roughly reflect the meaning of the given descriptions, but they fail to contain necessary details and vivid object parts. In this paper, we propose Stacked Generative Adversarial Networks (StackGAN) to generate 256×256 photo-realistic images conditioned on text descriptions. We decompose the hard problem into more manageable sub-problems through a sketch-refinement process. The Stage-I GAN sketches the primitive shape and colors of the object based on the given text description, yielding Stage-I low-resolution images. The Stage-II GAN takes Stage-I results and text descriptions as inputs, and generates high-resolution images with photo-realistic details. It is able to rectify defects in Stage-I results and add compelling details with the refinement process. To improve the diversity of the synthesized images and stabilize the training of the conditional-GAN, we introduce a novel Conditioning Augmentation technique that encourages smoothness in the latent conditioning manifold. Extensive experiments and comparisons with state-of-the-arts on benchmark datasets demonstrate that the proposed method achieves significant improvements on generating photo-realistic images conditioned on text descriptions.

Generative adversarial networks (GANs) are a recently proposed class of generative models in which a generator is trained to optimize a cost function that is being simultaneously learned by a discriminator. While the idea of learning cost functions is relatively new to the field of generative modeling, learning costs has long been studied in control and reinforcement learning (RL) domains, typically for imitation learning from demonstrations. In these fields, learning cost function underlying observed behavior is known as inverse reinforcement learning (IRL) or inverse optimal control.

While at first the connection between cost learning in RL and cost learning in generative modeling may appear to be a superficial one, we show in this paper that certain IRL methods are in fact mathematically equivalent to GANs. In particular, we demonstrate an equivalence between a sample-based algorithm for maximum entropy IRL and a GAN in which the generator’s density can be evaluated and is provided as an additional input to the discriminator. Interestingly, maximum entropy IRL is a special case of an energy-based model. We discuss the interpretation of GANs as an algorithm for training energy-based models, and relate this interpretation to other recent work that seeks to connect GANs and EBMs.

By formally highlighting the connection between GANs, IRL, and EBMs, we hope that researchers in all three communities can better identify and apply transferable ideas from one domain to another, particularly for developing more stable and scalable algorithms: a major challenge in all three domains.

Both generative adversarial networks (GAN) in unsupervised learning and actor-critic methods in reinforcement learning (RL) have gained a reputation for being difficult to optimize. Practitioners in both fields have amassed a large number of strategies to mitigate these instabilities and improve training. Here we show that GANs can be viewed as actor-critic methods in an environment where the actor cannot affect the reward. We review the strategies for stabilizing training for each class of models, both those that generalize between the two and those that are particular to that model. We also review a number of extensions to GANs and RL algorithms with even more complicated information flow. We hope that by highlighting this formal connection we will encourage both GAN and RL communities to develop general, scalable, and stable algorithms for multilevel optimization with deep networks, and to draw inspiration across communities.

The increasingly photorealistic sample quality of generative image models suggests their feasibility in applications beyond image generation. We present the Neural Photo Editor, an interface that leverages the power of generative neural networks to make large, semantically coherent changes to existing images. To tackle the challenge of achieving accurate reconstructions without loss of feature quality, we introduce the Introspective Adversarial Network, a novel hybridization of the VAE and GAN. Our model efficiently captures long-range dependencies through use of a computational block based on weight-shared dilated convolutions, and improves generalization performance with Orthogonal Regularization, a novel weight regularization method. We validate our contributions on CelebA, SVHN, and CIFAR-100, and produce samples and reconstructions with high visual fidelity.

Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4× upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.

This paper describes InfoGAN, an information-theoretic extension to the Generative Adversarial Network that is able to learn disentangled representations in a completely unsupervised manner. InfoGAN is a generative adversarial network that also maximizes the mutual information between a small subset of the latent variables and the observation. We derive a lower bound to the mutual information objective that can be optimized efficiently, and show that our training procedure can be interpreted as a variation of the Wake-Sleep algorithm.

Specifically, InfoGAN successfully disentangles writing styles from digit shapes on the MNIST dataset, pose from lighting of 3D rendered images, and background digits from the central digit on the SVHN dataset. It also discovers visual concepts that include hair styles, presence/​absence of eyeglasses, and emotions on the CelebA face dataset. Experiments show that InfoGAN learns interpretable representations that are competitive with representations learned by existing fully supervised methods.

We present a variety of new architectural features and training procedures that we apply to the generative adversarial networks (GANs) framework. We focus on two applications of GANs: semi-supervised learning, and the generation of images that humans find visually realistic. Unlike most work on generative models, our primary goal is not to train a model that assigns high likelihood to test data, nor do we require the model to be able to learn well without using any labels. Using our new techniques, we achieve state-of-the-art results in semi-supervised classification on MNIST, CIFAR-10 and SVHN. The generated images are of high quality as confirmed by a visual Turing test: our model generates MNIST samples that humans cannot distinguish from real data, and CIFAR-10 samples that yield a human error rate of 21.3%. We also present ImageNet samples with unprecedented resolution and show that our methods enable the model to learn recognizable features of ImageNet classes.

The ability of the Generative Adversarial Networks (GANs) framework to learn generative models mapping from simple latent distributions to arbitrarily complex data distributions has been demonstrated empirically, with compelling results showing that the latent space of such generators captures semantic variation in the data distribution. Intuitively, models trained to predict these semantic latent representations given data may serve as useful feature representations for auxiliary problems where semantics are relevant. However, in their existing form, GANs have no means of learning the inverse mapping—projecting data back into the latent space.

We propose Bidirectional Generative Adversarial Networks (BiGANs) as a means of learning this inverse mapping, and demonstrate that the resulting learned feature representation is useful for auxiliary supervised discrimination tasks, competitive with contemporary approaches to unsupervised and self-supervised feature learning.

In recent years, supervised learning with convolutional networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In this work we hope to help bridge the gap between the success of CNNs for supervised learning and unsupervised learning.

We introduce a class of CNNs called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning.

Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks—demonstrating their applicability as general image representations.