[video; Github; Colab] 3D avatar creation plays a crucial role in the digital age. However, the whole production process is prohibitively time-consuming and labor-intensive.
To democratize this technology to a larger audience, we propose AvatarCLIP, a zero-shot text-driven framework for 3D avatar generation and animation. Unlike professional software that requires expert knowledge, AvatarCLIP empowers layman users to customize a 3D avatar with the desired shape and texture, and drive the avatar with the described motions using solely natural languages.
Our key insight is to take advantage of the powerful vision-language model CLIP for supervising neural human generation, in terms of 3D geometry, texture and animation. Specifically, driven by natural language descriptions, we initialize 3D human geometry generation with a shape VAE network. Based on the generated 3D human shapes, a volume rendering model is utilized to further facilitate geometry sculpting and texture generation. Moreover, by leveraging the priors learned in the motion VAE, a CLIP-guided reference-based motion synthesis method is proposed for the animation of the generated 3D avatar.
Extensive qualitative and quantitative experiments validate the effectiveness and generalizability of AvatarCLIP on a wide range of avatars.
Remarkably, AvatarCLIP can generate unseen 3D avatars with novel animations, achieving superior zero-shot capability.
[samples] Text to speech (TTS) has made rapid progress in both academia and industry in recent years. Some questions naturally arise that whether a TTS system can achieve human-level quality, how to define/judge human-level quality and how to achieve it…Using this judge method, we found several previous TTS systems have not achieved it (see Table 1).
In this paper, we answer these questions by first defining human-level quality based on statistical-significance of measurement and describing the guidelines to judge it, and then proposing a TTS system called NaturalSpeech that achieves human-level quality on a benchmark dataset. Specifically, we leverage a variational autoencoder (VAE) for end-to-end text to waveform generation, with several key designs to enhance the capacity of prior from text and reduce the complexity of posterior from speech, including phoneme pre-training, differentiable duration modeling, bidirectional prior/posterior modeling, and memory mechanism in VAE.
Experiment evaluations on the popular LJSpeech dataset show that our proposed NaturalSpeech achieves -0.01 CMOS (comparative mean opinion score) to human recordings on sentence level, with Wilcoxon signed rank test at p-level p ≫0.05, which demonstrates no statistically-significant difference from human recordings for the first time on this dataset.
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.
Denoising diffusion probabilistic models are a promising new class of generative models that are competitive with GANs on perceptual metrics.
In this paper, we explore their potential for sequentially generating video. Inspired by recent advances in neural video compression, we use denoising diffusion models to stochastically generate a residual to a deterministic next-frame prediction.
We compare this approach to two sequential VAE and two GAN baselines on four datasets, where we test the generated frames for perceptual quality and forecasting accuracy against ground truth frames. We find substantial improvements in terms of perceptual quality on all data and improvements in terms of frame forecasting for complex high-resolution videos.
We present Polarity Sampling, a theoretically justified plug-and-play method for controlling the generation quality and diversity of pre-trained deep generative networks DGNs).
Leveraging the fact that DGNs are, or can be approximated by, continuous piecewise affine splines, we derive the analytical DGN output space distribution as a function of the product of the DGN’s Jacobian singular values raised to a power ρ. We dub ρ the polarity parameter and prove that ρ focuses the DGN sampling on the modes (ρ < 0) or anti-modes (ρ > 0) of the DGN output-space distribution.
We demonstrate that nonzero polarity values achieve a better precision-recall (quality-diversity) Pareto frontier than standard methods, such as truncation, for a number of state-of-the-art DGNs. We also present quantitative and qualitative results on the improvement of overall generation quality (eg. in terms of the Fréchet Inception Distance) for a number of state-of-the-art DGNs, including StyleGAN3, BigGAN-deep, NVAE, for different conditional and unconditional image generation tasks. In particular, Polarity Sampling redefines the state-of-the-art for StyleGAN2 on the FFHQ Dataset to FID 2.57, StyleGAN2 on the LSUN Car Dataset to FID 2.27 and StyleGAN3 on the AFHQv2 Dataset to FID 3.95.
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.
Variational autoencoders (VAE) are a popular approach to generative modelling. However, exploiting the capabilities of VAEs in practice can be difficult. Recent work on regularized and entropic autoencoders have begun to explore the potential, for generative modelling, of removing the variational approach and returning to the classic deterministic autoencoder (DAE) with additional novel regularization methods.
In this paper we empirically explore the capability of DAEs for image generation without additional novel methods and the effect of the implicit regularization and smoothness of large networks.
We find that DAEs can be used successfully for image generation without additional loss terms, and that many of the useful properties of VAEs can arise implicitly from sufficiently large convolutional encoders and decoders when trained on CIFAR-10 and CelebA.
We introduce CM3, a family of causally masked generative models trained over a large corpus of structured multi-modal documents that can contain both text and image tokens. Our new causally masked approach generates tokens left to right while also masking out a small number of long token spans that are generated at the end of the string, instead of their original positions. The casual masking object provides a type of hybrid of the more common causal and masked language models, by enabling full generative modeling while also providing bidirectional context when generating the masked spans. We train causally masked language-image models on large-scale web and Wikipedia articles, where each document contains all of the text, hypertext markup, hyperlinks, and image tokens (from a VQVAE-GAN), provided in the order they appear in the original HTML source (before masking). The resulting CM3 models can generate rich structured, multi-modal outputs while conditioning on arbitrary masked document contexts, and thereby implicitly learn a wide range of text, image, and cross modal tasks. They can be prompted to recover, in a zero-shot fashion, the functionality of models such as DALL·E, GENRE, and HTLM. We set the new state-of-the-art in zero-shot summarization, entity linking, and entity disambiguation while maintaining competitive performance in the fine-tuning setting. We can generate images unconditionally, conditioned on text (like DALL·E) and do captioning all in a zero-shot setting with a single model.
Diffusion Probabilistic models have been shown to generate state-of-the-art results on several competitive image synthesis benchmarks but lack a low-dimensional, interpretable latent space, and are slow at generation. On the other hand, Variational Autoencoders (VAEs) typically have access to a low-dimensional latent space but exhibit poor sample quality. Despite recent advances, VAEs usually require high-dimensional hierarchies of the latent codes to generate high-quality samples.
We present DiffuseVAE, a novel generative framework that integrates VAE within a diffusion model framework, and leverage this to design a novel conditional parameterization for diffusion models.
We show that the resulting model can improve upon the unconditional diffusion model in terms of sampling efficiency while also equipping diffusion models with the low-dimensional VAE inferred latent code. Furthermore, we show that the proposed model can generate high-resolution samples and exhibits synthesis quality comparable to state-of-the-art models on standard benchmarks. Lastly, we show that the proposed method can be used for controllable image synthesis and also exhibits out-of-the-box capabilities for downstream tasks like image super-resolution and denoising.
Conventional methods for the image-text generation tasks mainly tackle the naturally bidirectional generation tasks separately, focusing on designing task-specific frameworks to improve the quality and fidelity of the generated samples. Recently, Vision-Language Pre-training models have greatly improved the performance of the image-to-text generation tasks, but large-scale pre-training models for text-to-image synthesis task are still under-developed.
In this paper, we propose ERNIE-ViLG, an unified generative pre-training framework for bidirectional image-text generation with transformer model. Based on the image quantization models, we formulate both image generation and text generation as autoregressive generative tasks conditioned on the text/image input [cf. CogView, GLIDE, RuDOLPH, L-Verse, Huang et al 2021]. The bidirectional image-text generative modeling eases the semantic alignments across vision and language. For the text-to-image generation process, we further propose an end to end training method to jointly learn the visual sequence generator and the image reconstructor.
To explore the landscape of large-scale pre-training for bidirectional text-image generation, we train a 10-billion parameter ERNIE-ViLG model on a large-scale dataset of 145 million (Chinese) image-text pairs which achieves state-of-the-art performance for both text-to-image and image-to-text tasks, obtaining an FID of 7.9 on MS COCO for text-to-image synthesis and best results on COCO-CN and AIC-ICC for image captioning.
…The sources of our dataset are listed as follows:
Chinese Webpages: We crawl 800 million raw Chinese alt-text descriptions paired with images from various Chinese webpages, conduct several steps of filtering and totally harvest 70 million text-image pairs. The filtering rules mainly include: (1) Text-length: the number of words in alt-text is less than 15. (2) Text-content: the alt-text must contain at least one noun and contain no special characters. (3) Image-text similarity: the similarity score of between the alt-text and image (calculated by an in-house text-image matching model with the score range from 0.0 to 1.0) is greater than 0.5.
Image Search Engine: We collect roughly 60 million query texts and corresponding user-clicked images from our internal image search engine. There is often a strong correlation between the query and user-clicked images.
Public image-text Dataset: We collect a total of 15 million text-image pairs from two public datasets, CC30 and CC12M.40 The captions in these datasets are translated to Chinese through Baidu Translate API.
Figure 4: Example images ERNIE-ViLG generated in zero-shot setting with texts from open domain.Figure 4(a)–Figure 4(b) show the generated images of simple objects. Figure 4(c)–Figure 4(e) exhibit generated images of complex scenes with multiple objects. The example of creating image of non-existing objects is displayed in Figure 4(f).
Figure 5: Images of different styles generated by ERNIE-ViLG. “None” indicates not adding any prompts about image style.
Figure 6: Generated images given Chinese ancient poetry.
…Qualitative Results: ERNIE-ViLG has acquired the generation capability for various scenes, from basic objects to complex combinations of objects. Some examples are shown in Figure 4. As can be seen in the examples, ERNIE-ViLG can not only draw entities mentioned in the given text description, but also combine them together with the background in a reasonable way. Surprisingly, we also find two special skills ERNIE-ViLG develops. Firstly, ERNIE-ViLG can generate images of different styles by simply adding text prompts without fine-tuning like CogView does (Figure 5). Secondly, our model can generate realistic images given Chinese ancient poetry which shows promising understanding of brief and abstractive descriptions. Real concepts in the poetry are well-organized and artistic conception is well-described (Figure 6)
…We compare the results of our end-to-end training method with the two-stage pipeline baseline as shown in Table 7. For the two-stage pipeline, we train a text-to-image generator and use the decoder of dVAE directly as the reconstructor. The Two-stage G(R) refers to the separately trained generator(reconstructor), and the end-to-end G(R) refers to the end-to-end trained generator(reconstructor). Our end-to-end method achieves an important FID improvement of 1.5 compared to the two-stage pipeline. We find that combining the end-to-end trained generator (end-to-end G) and dVAE decoder (two-stage R) also brings a FID improvement of 0.9 compared to that of two-stage pipeline, but falls behind compared to the end-to-end methods. This indicates our proposed end-to-end method can improve both the performance of the generator (two-stage G & two-stage R vs end-to-end G & two-stage R) and the reconstructor (end-to-end G & two-stage R vs end-to-end G & end-to-end R).
We also input visual sequence of real images discretized by dVAE (gold image sequence) to the two reconstructors for comparison. Experimental results (the last two lines in Table 7) show that the end-to-end trained reconstructor has more obvious advantage in the reconstruction from real image discrete representation. We consider that end-to-end training will be more effective on ERNIE-ViLG with 10 billion parameters, for the image discrete representation generated by more capable generator is much closer to the real distribution, and hidden embeddings of larger model provides more useful features for the reconstructor. Due to the instability of the training of both GAN and large-scale generative model, we haven’t used end-to-end training for our 10-billion parameter model based on VQGAN. We will address the instability issue for future work and improve the 10-billion parameter ERNIE-ViLG through end-to-end training.
[Twitter samples; GLIDE; comparison] By decomposing the image formation process into a sequential application of denoising autoencoders, diffusion models (DMs) achieve state-of-the-art synthesis results on image data and beyond. Additionally, their formulation allows for a guiding mechanism to control the image generation process without retraining. However, since these models typically operate directly in pixel space, optimization of powerful DMs often consumes hundreds of GPU days and inference is expensive due to sequential evaluations.
To enable DM training on limited computational resources while retaining their quality and flexibility, we apply them in the latent space of powerful pretrained autoencoders. In contrast to previous work, training diffusion models on such a representation allows for the first time to reach a near-optimal point between complexity reduction and detail preservation, greatly boosting visual fidelity. By introducing cross-attention layers into the model architecture, we turn diffusion models into powerful and flexible generators for general conditioning inputs such as text or bounding boxes and high-resolution synthesis becomes possible in a convolutional manner…For a text-to-image model, we pre-train models conditioned on language prompts on the LAION database and finetune and evaluate on Conceptual Captions
Our latent diffusion models (LDMs) achieve a new state of the art for image inpainting and highly competitive performance on various tasks, including unconditional image generation, semantic scene synthesis, and super-resolution, while substantially reducing computational requirements compared to pixel-based DMs.
The integration of Vector Quantized Variational AutoEncoder (VQ-VAE) with autoregressive models as generation part has yielded high-quality results on image generation. However, the autoregressive models will strictly follow the progressive scanning order during the sampling phase. This leads the existing VQ series models to hardly escape the trap of lacking global information.
Denoising Diffusion Probabilistic Models (DDPM) in the continuous domain have shown a capability to capture the global context, while generating high-quality images. In the discrete state space, some works have demonstrated the potential to perform text generation and low resolution image generation. We show that with the help of a content-rich discrete visual codebook from VQ-VAE, the discrete diffusion model can also generate high fidelity images with global context, which compensates for the deficiency of the classical autoregressive model along pixel space.
Meanwhile, the integration of the discrete VAE with the diffusion model resolves the drawback of conventional autoregressive models being oversized, and the diffusion model which demands excessive time in the sampling process when generating images. It is found that the quality of the generated images is heavily dependent on the discrete visual codebook. Extensive experiments demonstrate that the proposed Vector Quantized Discrete Diffusion Model (VQ-DDM) is able to achieve comparable performance to top-tier methods with low complexity. It also demonstrates outstanding advantages over other vectors quantized with autoregressive models in terms of image inpainting tasks without additional training.
We present the vector quantized diffusion (VQ-Diffusion) model for text-to-image generation.
This method is based on a vector quantized variational autoencoder (VQ-VAE) whose latent space is modeled by a conditional variant of the recently developed Denoising Diffusion Probabilistic Model (DDPM). We find that this latent-space method is well-suited for text-to-image generation tasks because it not only eliminates the unidirectional bias with existing methods but also allows us to incorporate a mask-and-replace diffusion strategy to avoid the accumulation of errors, which is a serious problem with existing methods.
Our experiments show that the VQ-Diffusion produces substantially better text-to-image generation results when compared with conventional autoregressive (AR) models with similar numbers of parameters. Compared with previous GAN-based text-to-image methods, our VQ-Diffusion can handle more complex scenes and improve the synthesized image quality by a large margin. Finally, we show that the image generation computation in our method can be made highly efficient by reparameterization. With traditional AR methods, the text-to-image generation time increases linearly with the output image resolution and hence is quite time consuming even for normal size images. The VQ-Diffusion allows us to achieve a better trade-off between quality and speed.
Our experiments indicate that the VQ-Diffusion model with the reparameterization is 15× faster than traditional AR methods while achieving a better image quality.
Far beyond learning long-range interactions of natural language, transformers are becoming the de-facto standard for many vision tasks with their power and scalability. Especially with cross-modal tasks between image and text, vector quantized variational autoencoders (VQ-VAEs) are widely used to make a raw RGB image into a sequence of feature vectors. To better leverage the correlation between image and text, we propose L-Verse, a novel architecture consisting of feature-augmented variational autoencoder (AugVAE) and bidirectional auto-regressive transformer (BiART) for text-to-image and image-to-text generation.
Our AugVAE shows the state-of-the-art reconstruction performance on ImageNet1K validation set, along with the robustness to unseen images in the wild. Unlike other models, BiART can distinguish between image (or text) as a conditional reference and a generation target. L-Verse can be directly used for image-to-text or text-to-image generation tasks without any finetuning or extra object detection frameworks.
In quantitative and qualitative experiments, L-Verse shows impressive results against previous methods in both image-to-text and text-to-image generation on MS-COCO Captions. We furthermore assess the scalability of L-Verse architecture on Conceptual Captions and present the initial results of bidirectional vision-language representation learning on general domain.
Passive non-line-of-sight (NLOS) imaging has drawn great attention in recent years. However, all existing methods are in common limited to simple hidden scenes, low-quality reconstruction, and small-scale datasets.
In this paper, we propose NLOS-OT, a novel passive NLOS imaging framework based on manifold embedding and optimal transport, to reconstruct high-quality complicated hidden scenes. NLOS-OT converts the high-dimensional reconstruction task to a low-dimensional manifold mapping through optimal transport, alleviating the ill-posedness in passive NLOS imaging. Besides, we create the first large-scale passive NLOS imaging dataset, NLOS-Passive, which includes 50 groups and more than 3,200,000 images. NLOS-Passive collects target images with different distributions and their corresponding observed projections under various conditions, which can be used to evaluate the performance of passive NLOS imaging algorithms.
It is shown that the proposed NLOS-OT framework achieves much better performance than the state-of-the-art methods on NLOS-Passive.
We believe that the NLOS-OT framework together with the NLOS-Passive dataset is a big step and can inspire many ideas towards the development of learning-based passive NLOS imaging. Codes and dataset are publicly available.
[code] This paper shows that masked autoencoders (MAE) are scalable self-supervised learners for computer vision.
Our MAE approach is simple: we mask random patches of the input image and reconstruct the missing pixels. It is based on two core designs. First, we develop an asymmetric encoder-decoder architecture, with an encoder that operates only on the visible subset of patches (without mask tokens), along with a lightweight decoder that reconstructs the original image from the latent representation and mask tokens. Second, we find that masking a high proportion of the input image, eg. 75%, yields a nontrivial and meaningful self-supervisory task.
Coupling these two designs enables us to train large models efficiently and effectively: we accelerate training (by 3× or more) and improve accuracy. Our scalable approach allows for learning high-capacity models that generalize well: eg. a vanilla ViT-Huge model achieves the best accuracy (87.8%) among methods that use only ImageNet-1K data. Transfer performance in downstream tasks outperforms supervised pre-training and shows promising scaling behavior.
In order to better understand how the brain perceives faces, it is important to know what objective drives learning in the ventral visual stream.
To answer this question, we model neural responses to faces in the macaque inferotemporal (IT) cortex with a deep self-supervised generative model, β-VAE, which disentangles sensory data into interpretable latent factors, such as gender or age.
Our results demonstrate a strong correspondence between the generative factors discovered by β-VAE and those coded by single IT neurons, beyond that found for the baselines, including the handcrafted state-of-the-art model of face perception, the Active Appearance Model, and deep classifiers. Moreover, β-VAE is able to reconstruct novel face images using signals from just a handful of cells.
Together our results imply that optimizing the disentangling objective leads to representations that closely resemble those in the IT at the single unit level. This points at disentangling as a plausible learning objective for the visual brain.
Can visual artworks created using generative visual algorithms inspire human creativity in storytelling? We asked writers to write creative stories from a starting prompt, and provided them with visuals created by generative AI models from the same prompt. Compared to a control group, writers who used the visuals as story writing aid wrote statistically-significantly more creative, original, complete and visualizable stories, and found the task more fun. Of the generative algorithms used (BigGAN, VQGAN, DALL·E, CLIPDraw), VQGAN was the most preferred. The control group that did not view the visuals did significantly better in integrating the starting prompts. Findings indicate that cross modality inputs by AI can benefit divergent aspects of creativity in human-AI co-creation, but hinders convergent thinking.
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 72.2% for a similar model size. ViM-L also outperforms iGPT-XL which is trained with extra web image data and larger model size.
Score-based generative models (SGMs) have recently demonstrated impressive results in terms of both sample quality and distribution coverage. However, they are usually applied directly in data space and often require thousands of network evaluations for sampling.
Here, we propose the Latent Score-based Generative Model (LSGM), a novel approach that trains SGMs in a latent space, relying on the variational autoencoder framework. Moving from data to latent space allows us to train more expressive generative models, apply SGMs to non-continuous data, and learn smoother SGMs in a smaller space, resulting in fewer network evaluations and faster sampling.
To enable training LSGMs end-to-end in a scalable and stable manner, we (1) introduce a new score-matching objective suitable to the LSGM setting, (2) propose a novel parameterization of the score function that allows SGM to focus on the mismatch of the target distribution with respect to a simple Normal one, and (3) analytically derive multiple techniques for variance reduction of the training objective. LSGM obtains a state-of-the-art FID score of 2.10 on CIFAR-10, outperforming all existing generative results on this dataset. On CelebA-HQ-256, LSGM is on a par with previous SGMs in sample quality while outperforming them in sampling time by two orders of magnitude. In modeling binary images, LSGM achieves state-of-the-art likelihood on the binarized OMNIGLOT dataset.
Recent developments in the field of model-based RL have proven successful in a range of environments, especially ones where planning is essential. However, such successes have been limited to deterministic fully-observed environments. We present a new approach that handles stochastic and partially-observable environments. Our key insight is to use discrete autoencoders to capture the multiple possible effects of an action in a stochastic environment. We use a stochastic variant of Monte Carlo tree search to plan over both the agent’s actions and the discrete latent variables representing the environment’s response. Our approach significantly outperforms an offline version of MuZero on a stochastic interpretation of chess where the opponent is considered part of the environment. We also show that our approach scales to DeepMind Lab, a first-person 3D environment with large visual observations and partial observability.
In this work we introduce NWT, an expressive speech-to-video model. Unlike approaches that use domain-specific intermediate representations such as pose keypoints, NWT learns its own latent representations, with minimal assumptions about the audio and video content. To this end, we propose a novel discrete variational autoencoder with adversarial loss, dVAE-Adv, which learns a new discrete latent representation we call Memcodes. Memcodes are straightforward to implement, require no additional loss terms, are stable to train compared with other approaches, and show evidence of interpretability. To predict on the Memcode space, we use an autoregressive encoder-decoder model conditioned on audio. Additionally, our model can control latent attributes in the generated video that are not annotated in the data. We train NWT on clips from HBO’s Last Week Tonight with John Oliver. NWT consistently scores above other approaches in Mean Opinion Score (MOS) on tests of overall video naturalness, facial naturalness and expressiveness, and lipsync quality. This work sets a strong baseline for generalized audio-to-video synthesis. Samples are available at https://next-week-tonight.github.io/NWT/.
We present VideoGPT: a conceptually simple architecture for scaling likelihood based generative modeling to natural videos.
VideoGPT uses VQ-VAE that learns downsampled discrete latent representations of a raw video by employing 3D convolutions and axial self-attention. A simple GPT-like architecture is then used to autoregressively model the discrete latents using spatio-temporal position encodings.
Despite the simplicity in formulation and ease of training, our architecture is able to generate samples competitive with state-of-the-art GAN models for video generation on the BAIR Robot dataset, and generate high fidelity natural images from UCF101 and Tumbler GIF Dataset (TGIF).
We hope our proposed architecture serves as a reproducible reference for a minimalistic implementation of transformer based video generation models. Samples and code are available at https://wilson1yan.github.io/videogpt/index.html.
Score-based generative models and diffusion probabilistic models have been successful at generating high-quality samples in continuous domains such as images and audio. However, due to their Langevin-inspired sampling mechanisms, their application to discrete and sequential data has been limited.
In this work, we present a technique for training diffusion models on sequential data by parameterizing the discrete domain in the continuous latent space of a pre-trained variational autoencoder. Our method is non-autoregressive and learns to generate sequences of latent embeddings through the reverse process and offers parallel generation with a constant number of iterative refinement steps.
We apply this technique to modeling symbolic music and show strong unconditional generation and post-hoc conditional infilling results compared to autoregressive language models operating over the same continuous embeddings.
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.
Deep learning has enabled algorithms to generate realistic images. However, accurately predicting long video sequences requires understanding long-term dependencies and remains an open challenge. While existing video prediction models succeed at generating sharp images, they tend to fail at accurately predicting far into the future. We introduce the Clockwork VAE (CW-VAE), a video prediction model that leverages a hierarchy of latent sequences, where higher levels tick at slower intervals. We demonstrate the benefits of both hierarchical latents and temporal abstraction on 4 diverse video prediction datasets with sequences of up to 1000 frames, where CW-VAE outperforms top video prediction models. Additionally, we propose a Minecraft benchmark for long-term video prediction. We conduct several experiments to gain insights into CW-VAE and confirm that slower levels learn to represent objects that change more slowly in the video, and faster levels learn to represent faster objects.
Denoising diffusion probabilistic models (DDPMs) have achieved high quality image generation without adversarial training, yet they require simulating a Markov chain for many steps in order to produce a sample. To accelerate sampling, we present denoising diffusion implicit models (DDIMs), a more efficient class of iterative implicit probabilistic models with the same training procedure as DDPMs. In DDPMs, the generative process is defined as the reverse of a particular Markovian diffusion process. We generalize DDPMs via a class of non-Markovian diffusion processes that lead to the same training objective. These non-Markovian processes can correspond to generative processes that are deterministic, giving rise to implicit models that produce high quality samples much faster. We empirically demonstrate that DDIMs can produce high quality samples 10× to 50× faster in terms of wall-clock time compared to DDPMs, allow us to trade off computation for sample quality, perform semantically meaningful image interpolation directly in the latent space, and reconstruct observations with very low error.
[Paper: “DALL·E 1: Zero-Shot Text-to-Image Generation”, Ramesh et al 2021. Re-implementation: DALL·E 1 Mini (writeup). cf. CogView, Wu Dao. Availability through OA API still planned as of 2021-09-05; see DALL·E 2.] DALL·E 1 is a 12-billion parameter version of GPT-3 trained to generate images from text descriptions, using a dataset of text-image pairs. We’ve found that it has a diverse set of capabilities, including creating anthropomorphized versions of animals and objects, combining unrelated concepts in plausible ways, rendering text, and applying transformations to existing images.
GPT-3 showed that language can be used to instruct a large neural network to perform a variety of text generation tasks. iGPT showed that the same type of neural network can also be used to generate images with high fidelity. [iGPT is another answer to the question of “how do we do images autoregressively, but not at the exorbitant cost of generating pixels 1 by 1?”; iGPT uses ‘super pixels’ & very small images, while DALL·E 1 uses VAE ‘tokens’ corresponding roughly to small squares so the token sequence is relatively small, where the VAE does the actual compilation to raw pixels.] we extend these findings to show that manipulating visual concepts through language is now within reach.
[3 DALL·E 1 prompts: “an armchair in the shape of an avocado…” · “a store front that has the word ‘openai’ written on it…” · “the exact same cat on the top as a sketch on the bottom”]
DALL·E 1’s vocabulary has tokens for both text and image concepts. Specifically, each image caption is represented using a maximum of 256 BPE-encoded tokens with a vocabulary size of 16,384, and the image is represented using 1,024 tokens with a vocabulary size of 8,192. The images are preprocessed to 256×256 resolution during training. Similar to VQ-VAE,1415 each image is compressed to a 32×32 grid of discrete latent codes using a discrete VAE1011 that we pretrained using a continuous relaxation.1213 We found that training using the relaxation obviates the need for an explicit codebook, EMA loss, or tricks like dead code revival, and can scale up to large vocabulary sizes.
…Capabilities: We find that DALL·E 1 is able to create plausible images for a great variety of sentences that explore the compositional structure of language. We illustrate this using a series of interactive visuals in the next section. The samples shown for each caption in the visuals are obtained by taking the top 32 of 512 after reranking with CLIP [cf. CLIPScore, CogView’s caption scores], but we do not use any manual cherry-picking, aside from the thumbnails and standalone images that appear outside.
Controlling attributes: We test DALL·E 1’s ability to modify several of an object’s attributes, as well as the number of times that it appears.
Drawing multiple objects
Visualizing perspective and three-dimensionality
Visualizing internal and external structure
Inferring contextual details
We find that DALL·E 1 is able to render the same scene in a variety of different styles, and can adapt the lighting, shadows, and environment based on the time of day or season: “a … of a capybara sitting in a field at sunrise”
…With varying degrees of reliability, DALL·E 1 provides access to a subset of the capabilities of a 3D rendering engine via natural language. It can independently control the attributes of a small number of objects, and to a limited extent, how many there are, and how they are arranged with respect to one another. It can also control the location and angle from which a scene is rendered, and can generate known objects in compliance with precise specifications of angle and lighting conditions.
Zero-shot visual reasoning: GPT-3 can be instructed to perform many kinds of tasks solely from a description and a cue to generate the answer supplied in its prompt, without any additional training. For example, when prompted with the phrase “here is the sentence ‘a person walking his dog in the park’ translated into French:”, GPT-3 answers “un homme qui promène son chien dans le parc.” This capability is called zero-shot reasoning. We find that DALL·E 1 extends this capability to the visual domain, and is able to perform several kinds of image-to-image translation tasks when prompted in the right way. [See also CLIP.]
We did not anticipate that this capability would emerge, and made no modifications to the neural network or training procedure to encourage it. Motivated by these results, we measure DALL·E 1’s aptitude for analogical reasoning problems by testing it on Raven’s Progressive Matrices, a visual IQ test that saw widespread use in the 20th century. Rather than treating the IQ test a multiple-choice problem as originally intended, we ask DALL·E 1 to complete the bottom-right corner of each image using argmax sampling, and consider its completion to be correct if it is a close visual match to the original. DALL·E 1 is often able to solve matrices that involve continuing simple patterns or basic geometric reasoning, such as those in sets B and C. It is sometimes able to solve matrices that involve recognizing permutations and applying boolean operations, such as those in set D. The instances in set E tend to be the most difficult, and DALL·E 1 gets almost none of them correct. For each of the sets, we measure DALL·E 1’s performance on both the original images, and the images with the colors inverted. The inversion of colors should pose no additional difficulty for a human, yet does generally impair DALL·E 1’s performance, suggesting its capabilities may be brittle in unexpected ways.
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]
We combine the efficiency of convolutional approaches with the expressivity of transformers by introducing a convolutional VQGAN, which learns a codebook of context-rich visual parts, whose composition is modeled with an autoregressive transformer.
We present a hierarchical VAE that, for the first time, generates samples quickly while outperforming the PixelCNN in log-likelihood on all natural image benchmarks. We begin by observing that, in theory, VAEs can actually represent autoregressive models, as well as faster, better models if they exist, when made sufficiently deep. Despite this, autoregressive models have historically outperformed VAEs in log-likelihood. We test if insufficient depth explains why by scaling a VAE to greater stochastic depth than previously explored and evaluating it CIFAR-10, ImageNet, and FFHQ. In comparison to the PixelCNN, these very deep VAEs achieve higher likelihoods, use fewer parameters, generate samples thousands of times faster, and are more easily applied to high-resolution images. Qualitative studies suggest this is because the VAE learns efficient hierarchical visual representations. We release our source code and models at Github.
Normalizing flows, autoregressive models, variational autoencoders (VAEs), and deep energy-based models are among competing likelihood-based frameworks for deep generative learning. Among them, VAEs have the advantage of fast and tractable sampling and easy-to-access encoding networks. However, they are currently outperformed by other models such as normalizing flows and autoregressive models. While the majority of the research in VAEs is focused on the statistical challenges, we explore the orthogonal direction of carefully designing neural architectures for hierarchical VAEs. We propose Nouveau VAE (NVAE), a deep hierarchical VAE built for image generation using depth-wise separable convolutions and batch normalization. NVAE is equipped with a residual parameterization of Normal distributions and its training is stabilized by spectral regularization. We show that NVAE achieves state-of-the-art results among non-autoregressive likelihood-based models on the MNIST, CIFAR-10, CelebA 64, and CelebA HQ datasets and it provides a strong baseline on FFHQ. For example, on CIFAR-10, NVAE pushes the state-of-the-art from 2.98 to 2.91 bits per dimension, and it produces high-quality images on CelebA HQ. To the best of our knowledge, NVAE is the first successful VAE applied to natural images as large as 256×256 pixels. The source code is available at Github.
[Paper; samples; followup paper probing Jukebox as pretraining for music analysis (posing similar difficulties in extracting the right embedding as iGPT). An album made using it is Shadow Planet] A typical 4-minute song at CD quality (44 kHz, 16-bit) has over 10 million timesteps. For comparison, GPT-2 had 1,000 timesteps and OpenAI Five took tens of thousands of timesteps per game. Thus, to learn the high level semantics of music, a model would have to deal with extremely long-range dependencies. One way of addressing the long input problem is to use an autoencoder that compresses raw audio to a lower-dimensional space by discarding some of the perceptually irrelevant bits of information. We can then train a model to generate audio in this compressed space, and upsample back to the raw audio space.
We chose to work on music because we want to continue to push the boundaries of generative models. Our previous work on MuseNet explored synthesizing music based on large amounts of MIDI data. Now in raw audio, our models must learn to tackle high diversity as well as very long range structure, and the raw audio domain is particularly unforgiving of errors in short, medium, or long term timing.
…Jukebox’s autoencoder model compresses audio to a discrete space, using a quantization-based approach called VQ-VAE. Hierarchical VQ-VAEs can generate short instrumental pieces from a few sets of instruments, however they suffer from hierarchy collapse due to use of successive encoders coupled with autoregressive decoders. A simplified variant called VQ-VAE-226 avoids these issues by using feedforward encoders and decoders only, and they show impressive results at generating high-fidelity images…We use three levels in our VQ-VAE, shown below, which compress the 44kHz raw audio by 8×, 32×, and 128×, respectively, with a codebook size of 2,048 for each level. This downsampling loses much of the audio detail, and sounds noticeably noisy as we go further down the levels. However, it retains essential information about the pitch, timbre, and volume of the audio.
Jukebox architecture
…The top-level prior models the long-range structure of music, and samples decoded from this level have lower audio quality but capture high-level semantics like singing and melodies. The middle and bottom upsampling priors add local musical structures like timbre, substantially improving the audio quality. We train these as autoregressive models using a simplified variant of Sparse Transformers. Each of these models has 72 layers of factorized self-attention on a context of 8192 codes, which corresponds to ~24 seconds, 6 seconds, and 1.5 seconds of raw audio at the top, middle and bottom levels, respectively. Once all of the priors are trained, we can generate codes from the top level, upsample them using the upsamplers, and decode them back to the raw audio space using the VQ-VAE decoder to sample novel songs.
…While Jukebox represents a step forward in musical quality, coherence, length of audio sample, and ability to condition on artist, genre, and lyrics, there is a substantial gap between these generations and human-created music. For example, while the generated songs show local musical coherence, follow traditional chord patterns, and can even feature impressive solos, we do not hear familiar larger musical structures such as choruses that repeat. Our downsampling and upsampling process introduces discernible noise. Improving the VQ-VAE so its codes capture more musical information would help reduce this. Our models are also slow to sample from, because of the autoregressive nature of sampling. It takes approximately 9 hours to fully render 1 minute of audio through our models, and thus they cannot yet be used in interactive applications.
We introduce Jukebox, a model that generates music with singing in the raw audio domain.
We tackle the long context of raw audio using a multi-scale VQ-VAE to compress it to discrete codes, and modeling those using autoregressive Transformers. We show that the combined model at scale can generate high-fidelity and diverse songs with coherence up to multiple minutes.
We can condition on artist and genre to steer the musical and vocal style, and on unaligned lyrics to make the singing more controllable.
We are releasing thousands of non cherry-picked samples, along with model weights and code.
In the real world, RL agents should be rewarded for fulfilling human preferences. We show that RL agents implicitly learn the preferences of humans in their environment. Training a classifier to predict if a simulated human’s preferences are fulfilled based on the activations of a RL agent’s neural network gets .93 AUC. Training a classifier on the raw environment state gets only .8 AUC. Training the classifier off of the RL agent’s activations also does much better than training off of activations from an autoencoder. The human preference classifier can be used as the reward function of an RL agent to make RL agent more beneficial for humans.
We consider the problem of learning high-level controls over the global structure of generated sequences, particularly in the context of symbolic music generation with complex language models. In this work, we present the Transformer autoencoder, which aggregates encodings of the input data across time to obtain a global representation of style from a given performance. We show it is possible to combine this global representation with other temporally distributed embeddings, enabling improved control over the separate aspects of performance style and melody. Empirically, we demonstrate the effectiveness of our method on various music generation tasks on the MAESTRO dataset and a YouTube dataset with 10,000+ hours of piano performances, where we achieve improvements in terms of log-likelihood and mean listening scores as compared to baselines.
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.
We present BART, a denoising autoencoder for pretraining sequence-to-sequence models. BART is trained by (1) corrupting text with an arbitrary noising function, and (2) learning a model to reconstruct the original text. It uses a standard Tranformer-based neural machine translation architecture which, despite its simplicity, can be seen as generalizing BERT (due to the bidirectional encoder), GPT (with the left-to-right decoder), and many other more recent pretraining schemes. We evaluate a number of noising approaches, finding the best performance by both randomly shuffling the order of the original sentences and using a novel in-filling scheme, where spans of text are replaced with a single mask token. BART is particularly effective when fine tuned for text generation but also works well for comprehension tasks. It matches the performance of RoBERTa with comparable training resources on GLUE and SQuAD, achieves new state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains of up to 6 ROUGE. BART also provides a 1.1 BLEU increase over a back-translation system for machine translation, with only target language pretraining. We also report ablation experiments that replicate other pretraining schemes within the BART framework, to better measure which factors most influence end-task performance.
Gravitational wave (GW) detection is now commonplace and as the sensitivity of the global network of GW detectors improves, we will observe 𝒪(100)s of transient GW events per year. The current methods used to estimate their source parameters employ optimally sensitive but computationally costly Bayesian inference approaches where typical analyses have taken between 6 hours and 5 days. For binary neutron star and neutron star black hole systems prompt counterpart electromagnetic (EM) signatures are expected on timescales of 1s–1min and the current fastest method for alerting EM follow-up observers, can provide estimates in 𝒪(1) minutes, on a limited range of key source parameters.
Here we show that a conditional variational autoencoder pre-trained on binary black hole signals can return Bayesian posterior probability estimates.
The training procedure need only be performed once for a given prior parameter space and the resulting trained machine can then generate samples describing the posterior distribution ~6 orders of magnitude faster than existing techniques.
Collecting useful, interpretable, and biologically relevant phenotypes in a resource-efficient manner is a bottleneck to plant breeding, genetic mapping, and genomic prediction. Autonomous and affordable sub-canopy rovers are an efficient and scalable way to generate sensor-based datasets of in-field crop plants. Rovers equipped with light detection and ranging (LiDar) can produce three-dimensional reconstructions of entire hybrid maize fields. In this study, we collected 2,103 LiDar scans of hybrid maize field plots and extracted phenotypic data from them by Latent Space Phenotyping (LSP). We performed LSP by two methods, principal component analysis (PCA) and a convolutional autoencoder, to extract meaningful, quantitative Latent Space Phenotypes (LSPs) describing whole-plant architecture and biomass distribution. The LSPs had heritabilities of up to 0.44, similar to some manually measured traits, indicating they can be selected on or genetically mapped. Manually measured traits can be successfully predicted by using LSPs as explanatory variables in partial least squares regression, indicating the LSPs contain biologically relevant information about plant architecture. These techniques can be used to assess crop architecture at a reduced cost and in an automated fashion for breeding, research, or extension purposes, as well as to create or inform crop growth models.
We explore the use of Vector Quantized Variational AutoEncoder (VQ-VAE) models for large scale image generation. To this end, we scale and enhance the autoregressive priors used in VQ-VAE to generate synthetic samples of much higher coherence and fidelity than possible before. We use simple feed-forward encoder and decoder networks, making our model an attractive candidate for applications where the encoding and/or decoding speed is critical. Additionally, VQ-VAE requires sampling an autoregressive model only in the compressed latent space, which is an order of magnitude faster than sampling in the pixel space, especially for large images. We demonstrate that a multi-scale hierarchical organization of VQ-VAE, augmented with powerful priors over the latent codes, is able to generate samples with quality that rivals that of state of the art Generative Adversarial Networks on multifaceted datasets such as ImageNet, while not suffering from GAN’s known shortcomings such as mode collapse and lack of diversity.
Compilation of failed neural network experiments in generating anime images, pre-StyleGAN/BigGAN.
My experiments in generating anime faces, tried periodically since 2015, succeeded in 2019 with the release of StyleGAN. But for comparison, here are the failures from some of my older GAN or other NN attempts; as the quality is worse than StyleGAN, I won’t bother going into details—creating the datasets & training the ProGAN & tuning & transfer-learning were all much the same as already outlined at length for the StyleGAN results.
Included are:
ProGAN
Glow
MSG-GAN
PokeGAN
Self-Attention-GAN-TensorFlow
VGAN
BigGAN unofficial (official BigGAN is covered elsewhere)
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.
The gradient noise scale (appropriately averaged over training) explains the vast majority (R2 = 80%) of the variation in critical batch size over a range of tasks spanning six orders of magnitude. Batch sizes are measured in either number of images, tokens (for language models), or observations (for games).
…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.
Increasing parallelism makes it possible to train more complex models in a reasonable amount of time. We find that a Pareto frontier chart is the most intuitive way to visualize comparisons between algorithms and scales.
…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.
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 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.
Realistic music generation is a challenging task. When building generative models of music that are learnt from data, typically high-level representations such as scores or MIDI are used that abstract away the idiosyncrasies of a particular performance. But these nuances are very important for our perception of musicality and realism, so in this work we embark on modelling music in the raw audio domain.
It has been shown that autoregressive models excel at generating raw audio waveforms of speech, but when applied to music, we find them biased towards capturing local signal structure at the expense of modelling long-range correlations. This is problematic because music exhibits structure at many different timescales. In this work, we explore autoregressive discrete autoencoders (ADAs) as a means to enable autoregressive models to capture long-range correlations in waveforms. We find that they allow us to unconditionally generate piano music directly in the raw audio domain, which shows stylistic consistency across tens of seconds.
Learning a set of tasks over time, also known as continual learning (CL), is one of the most challenging problems in artificial intelligence. While recent approaches achieve some degree of CL in deep neural networks, they either (1) grow the network parameters linearly with the number of tasks, (2) require storing training data from previous tasks, or (3) restrict the network’s ability to learn new tasks.
To address these issues, we propose a novel framework, Self-Net, that uses an autoencoder to learn a set of low-dimensional representations of the weights learned for different tasks. We demonstrate that these low-dimensional vectors can then be used to generate high-fidelity recollections of the original weights. Self-Net can incorporate new tasks over time with little retraining and with minimal loss in performance for older tasks. Our system does not require storing prior training data and its parameters grow only logarithmically with the number of tasks.
We show that our technique outperforms current state-of-the-art approaches on numerous datasets—including continual versions of MNIST, CIFAR10, CIFAR100, and Atari—and we demonstrate that our method can achieve over 10× storage compression in a continual fashion.
To the best of our knowledge, we are the first to use autoencoders to sequentially encode sets of network weights to enable continual learning.
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.
Learning useful representations without supervision remains a key challenge in machine learning.
In this paper, we propose a simple yet powerful generative model that learns such discrete representations. Our model, the Vector Quantized-Variational AutoEncoder (VQ-VAE), differs from VAEs in two key ways: the encoder network outputs discrete, rather than continuous, codes; and the prior is learnt rather than static. In order to learn a discrete latent representation, we incorporate ideas from vector quantisation (VQ). Using the VQ method allows the model to circumvent issues of “posterior collapse”—where the latents are ignored when they are paired with a powerful autoregressive decoder—typically observed in the VAE framework.
Pairing these representations with an autoregressive prior, the model can generate high quality images, videos, and speech as well as doing high quality speaker conversion and unsupervised learning of phonemes, providing further evidence of the utility of the learnt representations.
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.
We introduce beta-VAE, a new state-of-the-art framework for automated discovery of interpretable factorised latent representations from raw image data in a completely unsupervised manner.
Learning an interpretable factorised representation of the independent data generative factors of the world without supervision is an important precursor for the development of artificial intelligence that is able to learn and reason in the same way that humans do. We introduce beta-VAE, a new state-of-the-art framework for automated discovery of interpretable factorised latent representations from raw image data in a completely unsupervised manner. Our approach is a modification of the variational autoencoder (VAE) framework. We introduce an adjustable hyperparameter beta that balances latent channel capacity and independence constraints with reconstruction accuracy. We demonstrate that beta-VAE with appropriately tuned beta > 1 qualitatively outperforms VAE (beta = 1), as well as state of the art unsupervised (InfoGAN) and semi-supervised (DC-IGN) approaches to disentangled factor learning on a variety of datasets (celebA, faces and chairs). Furthermore, we devise a protocol to quantitatively compare the degree of disentanglement learnt by different models, and show that our approach also significantly outperforms all baselines quantitatively. Unlike InfoGAN, beta-VAE is stable to train, makes few assumptions about the data and relies on tuning a single hyperparameter, which can be directly optimized through a hyper parameter search using weakly labeled data or through heuristic visual inspection for purely unsupervised data.
Generative models in vision have seen rapid progress due to algorithmic improvements and the availability of high-quality image datasets. In this paper, we offer contributions in both these areas to enable similar progress in audio modeling. First, we detail a powerful new WaveNet-style autoencoder model that conditions an autoregressive decoder on temporal codes learned from the raw audio waveform. Second, we introduce NSynth, a large-scale and high-quality dataset of musical notes that is an order of magnitude larger than comparable public datasets. Using NSynth, we demonstrate improved qualitative and quantitative performance of the WaveNet autoencoder over a well-tuned spectral autoencoder baseline. Finally, we show that the model learns a manifold of embeddings that allows for morphing between instruments, meaningfully interpolating in timbre to create new types of sounds that are realistic and expressive.
We introduce a method for learning the dynamics of complex nonlinear systems based on deep generative models over temporal segments of states and actions. Unlike dynamics models that operate over individual discrete timesteps, we learn the distribution over future state trajectories conditioned on past state, past action, and planned future action trajectories, as well as a latent prior over action trajectories. Our approach is based on convolutional autoregressive models and variational autoencoders. It makes stable and accurate predictions over long horizons for complex, stochastic systems, effectively expressing uncertainty and modeling the effects of collisions, sensory noise, and action delays. The learned dynamics model and action prior can be used for end-to-end, fully differentiable trajectory optimization and model-based policy optimization, which we use to evaluate the performance and sample-efficiency of our method.
Our world can be succinctly and compactly described as structured scenes of objects and relations. A typical room, for example, contains salient objects such as tables, chairs and books, and these objects typically relate to each other by their underlying causes and semantics. This gives rise to correlated features, such as position, function and shape. Humans exploit knowledge of objects and their relations for learning a wide spectrum of tasks, and more generally when learning the structure underlying observed data.
In this work, we introduce relation networks (RNs)—a general purpose neural network architecture for object-relation reasoning. We show that RNs are capable of learning object relations from scene description data. Furthermore, we show that RNs can act as a bottleneck that induces the factorization of objects from entangled scene description inputs, and from distributed deep representations of scene images provided by a variational autoencoder. The model can also be used in conjunction with differentiable memory mechanisms for implicit relation discovery in one-shot learning tasks. Our results suggest that relation networks are a potentially powerful architecture for solving a variety of problems that require object relation reasoning.
Categorical variables are a natural choice for representing discrete structure in the world. However, stochastic neural networks rarely use categorical latent variables due to the inability to backpropagate through samples. In this work, we present an efficient gradient estimator that replaces the non-differentiable sample from a categorical distribution with a differentiable sample from a novel Gumbel-Softmax distribution. This distribution has the essential property that it can be smoothly annealed into a categorical distribution. We show that our Gumbel-Softmax estimator outperforms state-of-the-art gradient estimators on structured output prediction and unsupervised generative modeling tasks with categorical latent variables, and enables large speedups on semi-supervised classification.
The reparameterization trick enables optimizing large scale stochastic computation graphs via gradient descent. The essence of the trick is to refactor each stochastic node into a differentiable function of its parameters and a random variable with fixed distribution. After refactoring, the gradients of the loss propagated by the chain rule through the graph are low variance unbiased estimators of the gradients of the expected loss. While many continuous random variables have such reparameterizations, discrete random variables lack useful reparameterizations due to the discontinuous nature of discrete states. In this work we introduce Concrete random variables—continuous relaxations of discrete random variables. The Concrete distribution is a new family of distributions with closed form densities and a simple reparameterization. Whenever a discrete stochastic node of a computation graph can be refactored into an one-hot bit representation that is treated continuously, Concrete stochastic nodes can be used with automatic differentiation to produce low-variance biased gradients of objectives (including objectives that depend on the log-probability of latent stochastic nodes) on the corresponding discrete graph. We demonstrate the effectiveness of Concrete relaxations on density estimation and structured prediction tasks using neural networks.
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.
Automated discovery of early visual concepts from raw image data is a major open challenge in AI research. Addressing this problem, we propose an unsupervised approach for learning disentangled representations of the underlying factors of variation. We draw inspiration from neuroscience, and show how this can be achieved in an unsupervised generative model by applying the same learning pressures as have been suggested to act in the ventral visual stream in the brain. By enforcing redundancy reduction, encouraging statistical independence, and exposure to data with transform continuities analogous to those to which human infants are exposed, we obtain a variational autoencoder (VAE) framework capable of learning disentangled factors. Our approach makes few assumptions and works well across a wide variety of datasets. Furthermore, our solution has useful emergent properties, such as zero-shot inference and an intuitive understanding of “objectness”.
We propose ways to improve the performance of fully connected networks. We found that two approaches in particular have a strong effect on performance: linear bottleneck layers and unsupervised pre-training using autoencoders without hidden unit biases. We show how both approaches can be related to improving gradient flow and reducing sparsity in the network.
We show that a fully connected network can yield ~70% classification accuracy on the permutation-invariant CIFAR-10 task, which is much higher than the current state-of-the-art. By adding deformations to the training data, the fully connected network achieves 78% accuracy, which is just 10% short of a decent convolutional network.
We present two approaches that use unlabeled data to improve sequence learning with recurrent networks. The first approach is to predict what comes next in a sequence, which is a conventional language model in natural language processing. The second approach is to use a sequence autoencoder, which reads the input sequence into a vector and predicts the input sequence again. These two algorithms can be used as a “pretraining” step for a later supervised sequence learning algorithm. In other words, the parameters obtained from the unsupervised step can be used as a starting point for other supervised training models. In our experiments, we find that long short term memory recurrent networks after being pretrained with the two approaches are more stable and generalize better. With pretraining, we are able to train long short term memory recurrent networks up to a few hundred timesteps, thereby achieving strong performance in many text classification tasks, such as IMDB, DBpedia and 20 Newsgroups.
How can we perform efficient inference and learning in directed probabilistic models, in the presence of continuous latent variables with intractable posterior distributions, and large datasets?
We introduce a stochastic variational inference and learning algorithm that scales to large datasets and, under some mild differentiability conditions, even works in the intractable case. Our contributions is two-fold. First, we show that a reparameterization of the variational lower bound yields a lower bound estimator that can be straightforwardly optimized using standard stochastic gradient methods. Second, we show that for i.i.d. datasets with continuous latent variables per datapoint, posterior inference can be made especially efficient by fitting an approximate inference model (also called a recognition model) to the intractable posterior using the proposed lower bound estimator.
Theoretical advantages are reflected in experimental results.
It defines a proper probabilistic model for the denoising autoencoder technique, which makes it in principle possible to sample from them or rank examples by their energy.
It suggests a different way to apply score matching [Hyvärinen 2005, Hyvärinen 2008] that is related to learning to denoise and does not require computing second derivatives.
It justifies the use of tied weights between the encoder and decoder and suggests ways to extend the success of denoising autoencoders to a larger family of energy-based models.
![[3 DALL·E 1 prompts: “an armchair in the shape of an avocado…” · “a store front that has the word ‘openai’ written on it…” · “the exact same cat on the top as a sketch on the bottom”]](/images/ai/2021-openai-dalle-textpromptexamples.png)
