Existing text-to-image (T2I) diffusion models face several limitations, including large model sizes, slow runtime, and low-quality generation on mobile devices. This paper aims to address all of these challenges by developing an extremely small and fast T2I model that generates high-resolution and high-quality images on mobile platforms. We propose several techniques to achieve this goal. First, we systematically examine the design choices of the network architecture to reduce model parameters and latency, while ensuring high-quality generation. Second, to further improve generation quality, we employ cross-architecture knowledge distillation from a much larger model, using a multi-level approach to guide the training of our model from scratch. Third, we enable a few-step generation by integrating adversarial guidance with knowledge distillation. For the first time, our model SnapGen, demonstrates the generation of 10242px images on a mobile device around 1.4 seconds. On ImageNet-1K, our model, with only 372M parameters, achieves an FID of 2.06 for 2562px generation. On T2I benchmarks (i.e., GenEval and DPG-Bench), our model with merely 379M parameters, surpasses large-scale models with billions of parameters at a significantly smaller size (e.g., 7× smaller than SDXL, 14× smaller than IF-XL).

How can one efficiently generate high-quality, wide-scope 3D scenes from arbitrary single images? Existing methods suffer several drawbacks, such as requiring multi-view data, time-consuming per-scene optimization, distorted geometry in occluded areas, and low visual quality in backgrounds. Our novel 3D scene reconstruction pipeline overcomes these limitations to tackle the aforesaid challenge. Specifically, we introduce a large-scale reconstruction model that leverages latents from a video diffusion model to predict 3D Gaussian Splattings of scenes in a feed-forward manner. The video diffusion model is designed to create videos precisely following specified camera trajectories, allowing it to generate compressed video latents that encode multi-view information while maintaining 3D consistency. We train the 3D reconstruction model to operate on the video latent space with a progressive learning strategy, enabling the efficient generation of high-quality, wide-scope, and generic 3D scenes. Extensive evaluations across various datasets affirm that our model significantly outperforms existing single-view 3D scene generation methods, especially with out-of-domain images. Thus, we demonstrate for the first time that a 3D reconstruction model can effectively be built upon the latent space of a diffusion model in order to realize efficient 3D scene generation. Project page: https://snap-research.github.io/wonderland/

We have witnessed the unprecedented success of diffusion-based video generation over the past year. Recently proposed models from the community have wielded the power to generate cinematic and high-resolution videos with smooth motions from arbitrary input prompts. However, as a supertask of image generation, video generation models require more computation and are thus hosted mostly on cloud servers, limiting broader adoption among content creators. In this work, we propose a comprehensive acceleration framework to bring the power of the large-scale video diffusion model to the hands of edge users. From the network architecture scope, we initialize from a compact image backbone and search out the design and arrangement of temporal layers to maximize hardware efficiency. In addition, we propose a dedicated adversarial fine-tuning algorithm for our efficient model and reduce the denoising steps to 4. Our model, with only 0.6B parameters, can generate a 5-second video on an iPhone 16 PM within 5 seconds. Compared to server-side models that take minutes on powerful GPUs to generate a single video, we accelerate the generation by magnitudes while delivering on-par quality. Project page at https://snap-research.github.io/snapgen-v/.

Direct Preference Optimization (DPO) has emerged as a powerful approach to align text-to-image (T2I) models with human feedback. Unfortunately, successful application of DPO to T2I models requires a huge amount of resources to collect and label large-scale datasets, e.g., millions of generated paired images annotated with human preferences. In addition, these human preference datasets can get outdated quickly as the rapid improvements of T2I models lead to higher quality images. In this work, we investigate a scalable approach for collecting large-scale and fully synthetic datasets for DPO training. Specifically, the preferences for paired images are generated using a pre-trained reward function, eliminating the need for involving humans in the annotation process, greatly improving the dataset collection efficiency. Moreover, we demonstrate that such datasets allow averaging predictions across multiple models and collecting ranked preferences as opposed to pairwise preferences. Furthermore, we introduce RankDPO to enhance DPO-based methods using the ranking feedback. Applying RankDPO on SDXL and SD3-Medium models with our synthetically generated preference dataset "Syn-Pic" improves both prompt-following (on benchmarks like T2I-Compbench, GenEval, and DPG-Bench) and visual quality (through user studies). This pipeline presents a practical and scalable solution to develop better preference datasets to enhance the performance of text-to-image models.

Recent approaches representing 3D objects and scenes using Gaussian splats show increased rendering speed across a variety of platforms and devices. While rendering such representations is indeed extremely efficient, storing and transmitting them is often prohibitively expensive. To represent large-scale scenes, one often needs to store millions of 3D Gaussian, which can occupy up to gigabytes of storage. This creates a significant practical barrier, preventing widespread adoption on resource-constrained devices. In this work, we propose a new representation that dramatically reduces the hard drive footprint while featuring similar or improved quality when compared to the standard 3D Gaussian splats. This representation leverages the inherent feature sharing among splats in the close proximity using a hierarchical tree structure, with which only the parent splats need to be stored. We present a method for constructing tree structures from naturally unstructured point clouds. Additionally, we propose the adaptive tree manipulation to prune the redundant trees in the space, while spawn new ones from the significant children splats during the optimization process. On the benchmark datasets, we achieve 20x storage reduction in hard-drive footprint with improved fidelity compared to the vanilla 3DGS and 2-5x reduction compared to the exiting compact solutions. More importantly, we demonstrate the practical application of our method in real-world rendering on mobile devices and AR glasses.

We present a novel video generation framework that integrates 3-dimensional geometry and dynamic awareness. To achieve this, we augment 2D videos with 3D point trajectories and align them in pixel space. The resulting 3D-aware video dataset, PointVid, is then used to fine-tune a latent diffusion model, enabling it to track 2D objects with 3D Cartesian coordinates. Building on this, we regularize the shape and motion of objects in the video to eliminate undesired artifacts, e.g., non-physical deformation. Consequently, we enhance the quality of generated RGB videos and alleviate common issues like object morphing, which are prevalent in current video models due to a lack of shape awareness. With our 3D augmentation and regularization, our model is capable of handling contact-rich scenarios such as task-oriented videos, where 3D information is essential for perceiving shape and motion of interacting solids. Our method can be seamlessly integrated into existing video diffusion models to improve their visual plausibility.

The quality of the data and annotation upper-bounds the quality of a downstream model. While there exist large text corpora and image-text pairs, high-quality video-text data is much harder to collect. First of all, manual labeling is more time-consuming, as it requires an annotator to watch an entire video. Second, videos have a temporal dimension, consisting of several scenes stacked together, and showing multiple actions. Accordingly, to establish a video dataset with high-quality captions, we propose an automatic approach leveraging multimodal inputs, such as textual video description, subtitles, and individual video frames. Specifically, we curate 3.8M high-resolution videos from the publicly available HD-VILA-100M dataset. We then split them into semantically consistent video clips, and apply multiple cross-modality teacher models to obtain captions for each video. Next, we finetune a retrieval model on a small subset where the best caption of each video is manually selected and then employ the model in the whole dataset to select the best caption as the annotation. In this way, we get 70M videos paired with high-quality text captions. We dub the dataset as Panda-70M. We show the value of the proposed dataset on three downstream tasks: video captioning, video and text retrieval, and text-driven video generation. The models trained on the proposed data score substantially better on the majority of metrics across all the tasks.

We present SPAD, a novel approach for creating con-sistent multi-view images from text prompts or single images. To enable multi-view generation, we repurpose a pre-trained 2D diffusion model by extending its self-attention layers with cross-view interactions, and fine-tune it on a high quality subset of Objaverse. We find that a naive extension of the self-attention proposed in prior work (e.g., MV-Dream) leads to content copying between views. Therefore, we explicitly constrain the cross-view attention based on epipolar geometry. To further enhance 3D consistency, we utilize Plücker coordinates derived from camera rays and inject them as positional encoding. This enables SPAD to reason over spatial proximity in 3D well. Compared to concurrent works that can only generate views at fixed azimuth and elevation (e.g., MVDream, SyncDreamer), SPAD offers full camera control and achieves state-of-the-art results in novel view synthesis on unseen objects from the Objaverse and Google Scanned Objects datasets. Finally, we demon-strate that text-to-3D generation using SPAD prevents the multi-face Janus issue.

Diffusion-based text-to-image generative models, e.g., Stable Diffusion, have revolutionized the field of content generation, enabling significant advancements in areas like image editing and video synthesis. Despite their formidable capabilities, these models are not without their limitations. It is still challenging to synthesize an image that aligns well with the input text, and multiple runs with carefully crafted prompts are required to achieve satisfactory results. To mitigate these limitations, numerous studies have endeavored to fine-tune the pre-trained diffusion models, i.e., UNet, utilizing various technologies. Yet, amidst these efforts, a pivotal question of text-to-image diffusion model training has remained largely unexplored: Is it possible and feasible to fine-tune the text encoder to improve the performance of text-to-image diffusion models? Our findings reveal that, instead of replacing the CLIP text encoder used in Stable Diffusion with other large language models, we can enhance it through our proposed fine-tuning approach, TextCraftor, leading to substantial improvements in quantitative benchmarks and human assessments. Interestingly, our technique also empowers controllable image generation through the interpolation of different text encoders fine-tuned with various rewards. We also demonstrate that TextCraftor is orthogonal to UNet finetuning, and can be combined to further improve generative quality.

Contemporary models for generating images show remarkable quality and versatility. Swayed by these advantages, the research community repurposes them to generate videos. Since video content is highly redundant, we argue that naively bringing advances of image models to the video generation domain reduces motion fidelity, visual quality and impairs scalability. In this work, we build Snap Video, a video-first model that systematically addresses these challenges. To do that, we first extend the EDM framework to take into account spatially and temporally redundant pixels and naturally support video generation. Second, we show that a U-Net—a workhorse behind image generation—scales poorly when generating videos, requiring significant computational overhead. Hence, we propose a new transformer-based architecture that trains 3.31 times faster than U-Nets (and is ∼4.5 faster at inference). This allows us to efficiently train a text-to-video model with billions of parameters for the first time, reach state-of-the-art results on a number of benchmarks, and generate videos with substantially higher quality, temporal consistency, and motion complexity. The user studies showed that our model was favored by a large margin over the most recent methods.

Despite significant advances in large-scale text-to-image models, achieving hyper-realistic human image generation remains a desirable yet unsolved task. Existing models like Stable Diffusion and DALL·E 2 tend to generate human images with incoherent parts or unnatural poses. To tackle these challenges, our key insight is that human image is inherently structural over multiple granularities, from the coarse-level body skeleton to fine-grained spatial geometry. Therefore, capturing such correlations between the explicit appearance and latent structure in one model is essential to generate coherent and natural human images. To this end, we propose a unified framework, HyperHuman, that generates in-the-wild human images of high realism and diverse layouts. Specifically, 1) we first build a large-scale human-centric dataset, named HumanVerse, which consists of 340M images with comprehensive annotations like human pose, depth, and surface normal. 2) Next, we propose a Latent Structural Diffusion Model that simultaneously denoises the depth and surface normal along with the synthesized RGB image. Our model enforces the joint learning of image appearance, spatial relationship, and geometry in a unified network, where each branch in the model complements to each other with both structural awareness and textural richness. 3) Finally, to further boost the visual quality, we propose a Structure-Guided Refiner to compose the predicted conditions for more detailed generation of higher resolution. Extensive experiments demonstrate that our framework yields the state-of-the-art performance, generating hyper-realistic human images under diverse scenarios.

We present ``Magic123'', a two-stage coarse-to-fine approach for high-quality, textured 3D mesh generation from a single image in the wild using *both 2D and 3D priors*. In the first stage, we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually appealing texture. In both stages, the 3D content is learned through reference-view supervision and novel-view guidance by a joint 2D and 3D diffusion prior. We introduce a trade-off parameter between the 2D and 3D priors to control the details and 3D consistencies of the generation. Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated through extensive experiments on diverse synthetic and real-world images.

One highly promising direction for enabling flexible real-time on-device image editing is utilizing data distillation by leveraging large-scale text-to-image diffusion models to generate paired datasets used for training generative adversarial networks (GANs). This approach notably alleviates the stringent requirements typically imposed by high-end commercial GPUs for performing image editing with diffusion models. However, unlike text-to-image diffusion models, each distilled GAN is specialized for a specific image editing task, necessitating costly training efforts to obtain models for various concepts. In this work, we introduce and address a novel research direction: can the process of distilling GANs from diffusion models be made significantly more efficient? To achieve this goal, we propose a series of innovative techniques. First, we construct a base GAN model with generalized features, adaptable to different concepts through fine-tuning, eliminating the need for training from scratch. Second, we identify crucial layers within the base GAN model and employ Low-Rank Adaptation (LoRA) with a simple yet effective rank search process, rather than fine-tuning the entire base model. Third, we investigate the minimal amount of data necessary for fine-tuning, further reducing the overall training time. Extensive experiments show that we can efficiently empower GANs with the ability to perform real-time high-quality image editing on mobile devices with remarkably reduced training and storage costs for each concept.

Neural network architecture design requires making many crucial decisions. The common desiderata is that similar decisions, with little modifications, can be reused in a variety of tasks and applications. To satisfy that, architectures must provide promising latency and performance trade-offs, support a variety of tasks, scale efficiently with respect to the amounts of data and compute, leverage available data from other tasks, and efficiently support various hardware. To this end, we introduce AsCAN---a hybrid architecture, combining both convolutional and transformer blocks. We revisit the key design principles of hybrid architectures and propose a simple and effective \emph{asymmetric} architecture, where the distribution of convolutional and transformer blocks is \emph{asymmetric}, containing more convolutional blocks in the earlier stages, followed by more transformer blocks in later stages. AsCAN supports a variety of tasks: recognition, segmentation, class-conditional image generation, and features a superior trade-off between performance and latency. We then scale the same architecture to solve a large-scale text-to-image task and show state-of-the-art performance compared to the most recent public and commercial models. Notably, without performing any optimization of inference time our model shows faster execution, even when compared to works that do such optimization, highlighting the advantages and the value of our approach.

Diffusion-based image generation models have achieved great success in recent years by showing the capability of synthesizing high-quality content. However, these models contain a huge number of parameters, resulting in a significantly large model size. Saving and transferring them is a major bottleneck for various applications, especially those running on resource-constrained devices. In this work, we develop a novel weight quantization method that quantizes the UNet from Stable Diffusion v1.5 to $1.99$ bits, achieving a model with $7.9\times$ smaller size while exhibiting even better generation quality than the original one. Our approach includes several novel techniques, such as assigning optimal bits to each layer, initializing the quantized model for better performance, and improving the training strategy to dramatically reduce quantization error. Furthermore, we extensively evaluate our quantized model across various benchmark datasets and through human evaluation to demonstrate its superior generation quality.

Diffusion-based video generation models have demonstrated remarkable success in obtaining high-fidelity videos through the iterative denoising process. However, these models require multiple denoising steps during sampling, resulting in high computational costs. In this work, we propose a novel approach to obtain single-step video generation models by leveraging adversarial training to fine-tune pre-trained video diffusion models. We show that, through the adversarial training, the multi-steps video diffusion model, i.e., Stable Video Diffusion (SVD), can be trained to perform single forward pass to synthesize high-quality videos, capturing both temporal and spatial dependencies in the video data. Extensive experiments demonstrate that our method achieves competitive generation quality of synthesized videos with significantly reduced computational overhead for the denoising process (i.e., around $23\times$ speedup compared with SVD and $6\times$ speedup compared with existing works, with even better generation quality), paving the way for real-time video synthesis and editing.

Recent efforts in Neural Rendering Fields (NeRF) have shown impressive results on novel view synthesis by utilizing implicit neural representation to represent 3D scenes. Due to the process of volumetric rendering, the inference speed for NeRF is extremely slow, limiting the application scenarios of utilizing NeRF on resource-constrained hardware, such as mobile devices. Many works have been conducted to reduce the latency of running NeRF models. However, most of them still require high-end GPU for acceleration or extra storage memory, which is all unavailable on mobile devices. Another emerging direction utilizes the neural light field (NeLF) for speedup, as only one forward pass is performed on a ray to predict the pixel color. Nevertheless, to reach a similar rendering quality as NeRF, the network in NeLF is designed with intensive computation, which is not mobile-friendly. In this work, we propose an efficient network that runs in real-time on mobile devices for neural rendering. We follow the setting of NeLF to train our network. Unlike existing works, we introduce a novel network architecture that runs efficiently on mobile devices with low latency and small size, i.e., saving 15 x ~ 24 x storage compared with MobileNeRF. Our model achieves high-resolution generation while maintaining real-time inference for both synthetic and real-world scenes on mobile devices, e.g., 18.04ms (iPhone 13) for rendering one$1008 \times 756$image of real 3D scenes. Additionally, we achieve similar image quality as NeRF and better quality than MobileNeRF (PSNR 26.15 vs. 25.91 on the real-world forward-facing dataset)11More demo examples in our Webpage..

Building animatable and editable models of clothed humans from raw 3D scans and poses is a challenging problem. Existing reposing methods suffer from the limited expressiveness of Linear Blend Skinning (LBS), require costly mesh extraction to generate each new pose, and typically do not preserve surface correspondences across different poses. In this work, we introduce Invertible Neural Skinning (INS) to address these shortcomings. To maintain correspondences, we propose a Pose-conditioned Invertible Network (PIN) architecture, which extends the LBS process by learning additional pose-varying deformations. Next, we combine PIN with a differentiable LBS module to build an expressive and end-to-end Invertible Neural Skinning (INS) pipeline. We demonstrate the strong performance of our method by outperforming the state-of-the-art reposing techniques on clothed humans and preserving surface correspondences, while being an order of magnitude faster. We also perform an ablation study, which shows the usefulness of our pose-conditioning formulation, and our qualitative results display that INS can rectify artefacts introduced by LBS well.

There has been a recent explosion of impressive generative models that can produce high quality images (or videos) conditioned on text descriptions. However, all such approaches rely on conditional sentences that contain un-ambiguous descriptions of scenes and main actors in them. Therefore employing such models for more complex task of story visualization, where naturally references and co-references exist, and one requires to reason about when to maintain consistency of actors and backgrounds across frames/scenes, and when not to, based on story progression, remains a challenge. In this work, we address the aforementioned challenges and propose a novel autoregressive diffusion-based framework with a visual memory module that implicitly captures the actor and background context across the generated frames. Sentence-conditioned soft attention over the memories enables effective reference resolution and learns to maintain scene and actor consistency when needed. To validate the effectiveness of our approach, we extend the MUGEN dataset [19] and introduce additional characters, backgrounds and referencing in multi-sentence storylines. Our experiments for story generation on the MUGEN, the PororoSV [30] and the FlintstonesSV [16] dataset show that our method not only outperforms prior state-of-the-art in generating frames with high visual quality, which are consistent with the story, but also models appropriate correspondences between the characters and the background.

We propose a novel approach for unsupervised 3D animation of non-rigid deformable objects. Our method learns the 3D structure and dynamics of objects solely from single-view RGB videos, and can decompose them into semantically meaningful parts that can be tracked and animated. Using a 3D autodecoder framework, paired with a keypoint estimator via a differentiable PnP algorithm, our model learns the underlying object geometry and parts decomposition in an entirely unsupervised manner. This allows it to perform 3D segmentation, 3D keypoint estimation, novel view synthesis, and animation. We primarily evaluate the framework on two video datasets: VoxCeleb 2562and TEDXPeople 2562. In addition, on the Cats 2562image dataset, we show it even learns compelling 3D geometry from still images. Finally, we show our model can obtain animatable 3D objects from a single or few images11Code and visual results available on our project website: https://snap-research.github.io/unsupervised-volumetric-animation..

Recent works on diffusion models have demonstrated a strong capability for conditioning image generation, e.g., text-guided image synthesis. Such success inspires many efforts trying to use large-scale pre-trained diffusion models for tackling a challenging problem-real image editing. Works conducted in this area learn a unique textual token corresponding to several images containing the same object. However, under many circumstances, only one image is available, such as the painting of the Girl with a Pearl Earring. Using existing works on fine-tuning the pre-trained diffusion models with a single image causes severe overfitting issues. The information leakage from the pre-trained diffusion models makes editing can not keep the same content as the given image while creating new features depicted by the language guidance. This work aims to address the problem of single-image editing. We propose a novel model-based guidance built upon the classifier-free guidance so that the knowledge from the model trained on a single image can be distilled into the pre-trained diffusion model, enabling content creation even with one given image. Additionally, we propose a patch-based fine-tuning that can effectively help the model generate images of arbitrary resolution. We provide extensive experiments to validate the design choices of our approach and show promising editing capabilities, including changing style, content addition, and object manipulation. Our code is made publicly available here.

With the success of Vision Transformers (ViTs) in computer vision tasks, recent arts try to optimize the performance and complexity of ViTs to enable efficient deployment on mobile devices. Multiple approaches are proposed to accelerate attention mechanism, improve inefficient designs, or incorporate mobile-friendly lightweight convolutions to form hybrid architectures. However, ViT and its variants still have higher latency or considerably more parameters than lightweight CNNs, even true for the years-old MobileNet. In practice, latency and size are both crucial for efficient deployment on resource-constraint hardware. In this work, we investigate a central question, can transformer models run as fast as MobileNet and maintain a similar size? We revisit the design choices of ViTs and propose a novel supernet with low latency and high parameter efficiency. We further introduce a novel fine-grained joint search strategy for transformer models that can find efficient architectures by optimizing latency and number of parameters simultaneously. The proposed models, EfficientFormerV2, achieve 3.5% higher top-1 accuracy than MobileNetV2 on ImageNet-1K with similar latency and parameters. This work demonstrate that properly designed and optimized vision transformers can achieve high performance even with MobileNet-level size and speed1.

Attention-based vision models, such as Vision Transformer (ViT) and its variants, have shown promising performance in various computer vision tasks. However, these emerging architectures suffer from large model sizes and high computational costs, calling for efficient model compression solutions. To date, pruning ViTs has been well studied, while other compression strategies that have been widely applied in CNN compression, e.g., model factorization, is little explored in the context of ViT compression. This paper explores an efficient method for compressing vision transformers to enrich the toolset for obtaining compact attention-based vision models. Based on the new insight on the multi-head attention layer, we develop a highly efficient ViT compression solution, which outperforms the state-of-the-art pruning methods. For compressing DeiT-small and DeiT-base models on ImageNet, our proposed approach can achieve $0.45%$ and $0.76%$ higher top-1 accuracy even with fewer parameters. Our finding can also be applied to improve the customization efficiency of text-to-image diffusion models, with much faster training (up to $2.6\times$ speedup) and lower extra storage cost (up to $1927.5\times$ reduction) than the existing works.

Real-time novel-view image synthesis on mobile devices is prohibitive due to the limited computational power and storage. Using volumetric rendering methods, such as NeRF and its derivatives, on mobile devices is not suitable due to the high computational cost of volumetric rendering. On the other hand, recent advances in neural light field representations have shown promising real-time view synthesis results on mobile devices. Neural light field methods learn a direct mapping from a ray representation to the pixel color. The current choice of ray representation is either stratified ray sampling or Plücker coordinates, overlooking the classic light slab (two-plane) representation, the preferred representation to interpolate between light field views. In this work, we find that using the light slab representation is an efficient representation for learning a neural light field. More importantly, it is a lower-dimensional ray representation enabling us to learn the 4D ray space using feature grids which are significantly faster to train and render. Although mostly designed for frontal views, we show that the light-slab representation can be further extended to non-frontal scenes using a divide-and-conquer strategy. Our method provides better rendering quality than prior light field methods and a significantly better trade-off between rendering quality and speed than prior light field methods.

Text-to-image diffusion models can create stunning images from natural language descriptions that rival the work of professional artists and photographers. However, these models are large, with complex network architectures and tens of denoising iterations, making them computationally expensive and slow to run. As a result, high-end GPUs and cloud-based inference are required to run diffusion models at scale. This is costly and has privacy implications, especially when user data is sent to a third party. To overcome these challenges, we present a generic approach that, for the first time, unlocks running text-to-image diffusion models on mobile devices in **less than 2 seconds**. We achieve so by introducing efficient network architecture and improving step distillation. Specifically, we propose an efficient UNet by identifying the redundancy of the original model and reducing the computation of the image decoder via data distillation. Further, we enhance the step distillation by exploring training strategies and introducing regularization from classifier-free guidance. Our extensive experiments on MS-COCO show that our model with $8$ denoising steps achieves better FID and CLIP scores than Stable Diffusion v$1.5$ with $50$ steps. Our work democratizes content creation by bringing powerful text-to-image diffusion models to the hands of users.

In this paper, we present a method for generating consistent novel views from a single source image. Our approach focuses on maximizing the reuse of visible pixels from the source image. To achieve this, we use a monocular depth estimator that transfers visible pixels from the source view to the target view. Starting from a pre-trained 2D inpainting diffusion model, we train our method on the large-scale Objaverse dataset to learn 3D object priors. While training we use a novel masking mechanism based on epipolar lines to further improve the quality of our approach. This allows our framework to perform zero-shot novel view synthesis on a variety of objects. We evaluate the zero-shot abilities of our framework on three challenging datasets: Google Scanned Objects, Ray Traced Multiview, and Common Objects in 3D.

Multimodal data collected from the real world are often imperfect due to missing modalities. Therefore multimodal models that are robust against modal-incomplete data are highly preferred. Recently, Transformer models have shown great success in processing multimodal data. However, existing work has been limited to either architecture designs or pre-training strategies; whether Transformer models are naturally robust against missing-modal data has rarely been investigated. In this paper, we present the first-of-its-kind work to comprehensively investigate the behavior of Transformers in the presence of modal-incomplete data. Unsurprising, we find Transformer models are sensitive to missing modalities while different modal fusion strategies will significantly affect the robustness. What surprised us is that the optimal fusion strategy is dataset dependent even for the same Transformer model; there does not exist a universal strategy that works in general cases. Based on these findings, we propose a principle method to improve the robustness of Transformer models byautomatically searching for an optimal fusion strategy regarding input data. Experimental validations on three benchmarks support the superior performance of the proposed method.

Image outpainting seeks for a semantically consistent extension of the input image beyond its available content. Compared to inpainting - filling in missing pixels in a way coherent with the neighboring pixels - outpainting can be achieved in more diverse ways since the problem is less constrained by the surrounding pixels. Existing image outpainting methods pose the problem as a conditional image-to-image translation task, often generating repetitive structures and textures by replicating the content available in the input image. In this work, we formulate the problem from the perspective of inverting generative adversarial networks. Our generator renders micro-patches conditioned on their joint latent code as well as their individual positions in the image. To outpaint an image, we seek for multiple latent codes not only recovering available patches but also synthesizing diverse outpainting by patch-based generation. This leads to richer structure and content in the outpainted regions. Furthermore, our formulation allows for outpainting conditioned on the categorical input, thereby enabling flexible user controls. Extensive experimental results demonstrate the proposed method performs favorably against existing in- and outpainting methods, featuring higher visual quality and diversity.

Most methods for conditional video synthesis use a single modality as the condition. This comes with major limitations. For example, it is problematic for a model conditioned on an image to generate a specific motion trajectory desired by the user since there is no means to provide motion information. Conversely, language information can describe the desired motion, while not precisely defining the content of the video. This work presents a multimodal video generation framework that benefits from text and images provided jointly or separately. We leverage the recent progress in quantized representations for videos and apply a bidirectional transformer with multiple modalities as inputs to predict a discrete video representation. To improve video quality and consistency, we propose a new video token trained with self-learning and an improved mask-prediction algorithm for sampling video tokens. We introduce text augmentation to improve the robustness of the textual representation and diversity of generated videos. Our framework can incorporate various visual modalities, such as segmentation masks, drawings, and partially occluded images. It can generate much longer sequences than the one used for training. In addition, our model can extract visual information as suggested by the text prompt, e.g., “an object in image one is moving northeast”, and generate corresponding videos. We run evaluations on three public datasets and a newly collected dataset labeled with facial attributes, achieving state-of-the-art generation results on all four11Code: https://github.com/snap-research/MMVID and Webpage..

Vision Transformers (ViT) have shown rapid progress in computer vision tasks, achieving promising results on various benchmarks. However, due to the massive number of parameters and model design, e.g., attention mechanism, ViT-based models are generally times slower than lightweight convolutional networks. Therefore, the deployment of ViT for real-time applications is particularly challenging, especially on resource-constrained hardware such as mobile devices. Recent efforts try to reduce the computation complexity of ViT through network architecture search or hybrid design with MobileNet block, yet the inference speed is still unsatisfactory. This leads to an important question: can transformers run as fast as MobileNet while obtaining high performance? To answer this, we first revisit the network architecture and operators used in ViT-based models and identify inefficient designs. Then we introduce a dimension-consistent pure transformer (without MobileNet blocks) as a design paradigm. Finally, we perform latency-driven slimming to get a series of final models dubbed EfficientFormer. Extensive experiments show the superiority of EfficientFormer in performance and speed on mobile devices. Our fastest model, EfficientFormer-L1, achieves $79.2\%$ top-1 accuracy on ImageNet-1K with only $1.6$ ms inference latency on iPhone 12 (compiled with CoreML), which runs as fast as MobileNetV2$\times 1.4$ ($1.6$ ms, $74.7\%$ top-1), and our largest model, EfficientFormer-L7, obtains $83.3\%$ accuracy with only $7.0$ ms latency. Our work proves that properly designed transformers can reach extremely low latency on mobile devices while maintaining high performance.

Recently, sparse training has emerged as a promising paradigm for efficient deep learning on edge devices. The current research mainly devotes the efforts to reducing training costs by further increasing model sparsity. However, increasing sparsity is not always ideal since it will inevitably introduce severe accuracy degradation at an extremely high sparsity level. This paper intends to explore other possible directions to effectively and efficiently reduce sparse training costs while preserving accuracy. To this end, we investigate two techniques, namely, layer freezing and data sieving. First, the layer freezing approach has shown its success in dense model training and fine-tuning, yet it has never been adopted in the sparse training domain. Nevertheless, the unique characteristics of sparse training may hinder the incorporation of layer freezing techniques. Therefore, we analyze the feasibility and potentiality of using the layer freezing technique in sparse training and find it has the potential to save considerable training costs. Second, we propose a data sieving method for dataset-efficient training, which further reduces training costs by ensuring only a partial dataset is used throughout the entire training process. We show that both techniques can be well incorporated into the sparse training algorithm to form a generic framework, which we dub SpFDE. Our extensive experiments demonstrate that SpFDE can significantly reduce training costs while preserving accuracy from three dimensions: weight sparsity, layer freezing, and dataset sieving. Our code and models will be released.

A common assumption in multimodal learning is the completeness of training data, i.e., full modalities are available in all training examples. Although there exists research endeavor in developing novel methods to tackle the incompleteness of testing data, e.g., modalities are partially missing in testing examples, few of them can handle incomplete training modalities. The problem becomes even more challenging if considering the case of severely missing, e.g., ninety percent of training examples may have incomplete modalities. For the first time in the literature, this paper formally studies multimodal learning with missing modality in terms of flexibility (missing modalities in training, testing, or both) and efficiency (most training data have incomplete modality). Technically, we propose a new method named SMIL that leverages Bayesian meta-learning in uniformly achieving both objectives. To validate our idea, we conduct a series of experiments on three popular benchmarks: MM-IMDb, CMU-MOSI, and avMNIST. The results prove the state-of-the-art performance of SMIL over existing methods and generative baselines including autoencoders and generative adversarial networks.

Generative Adversarial Networks (GANs) have achieved huge success in generating high-fidelity images, however, they suffer from low efficiency due to tremendous computational cost and bulky memory usage. Recent efforts on compression GANs show noticeable progress in obtaining smaller generators by sacrificing image quality or involving a time-consuming searching process. In this work, we aim to address these issues by introducing a teacher network that provides a search space in which efficient network architectures can be found, in addition to performing knowledge distillation. First, we revisit the search space of generative models, introducing an inception-based residual block into generators. Second, to achieve target computation cost, we propose a one-step pruning algorithm that searches a student architecture from the teacher model and substantially reduces searching cost. It requires no ℓ1sparsity regularization and its associated hyper-parameters, simplifying the training procedure. Finally, we propose to distill knowledge through maximizing feature similarity between teacher and student via an index named Global Kernel Alignment (GKA). Our compressed networks achieve similar or even better image fidelity (FID, mIoU) than the original models with much-reduced computational cost, e.g., MACs. Code will be released at https://github.com/snap-research/CAT.

Human motion retargeting aims to transfer the motion of one person in a "driving" video or set of images to another person. Existing efforts leverage a long training video from each target person to train a subject-specific motion transfer model. However, the scalability of such methods is limited, as each model can only generate videos for the given target subject, and such training videos are labor-intensive to acquire and process. Few-shot motion transfer techniques, which only require one or a few images from a target, have recently drawn considerable attention. Methods addressing this task generally use either 2D or explicit 3D representations to transfer motion, and in doing so, sacrifice either accurate geometric modeling or the flexibility of an end-to-end learned representation. Inspired by the Transformable Bottleneck Network, which renders novel views and manipulations of rigid objects, we propose an approach based on an implicit volumetric representation of the image content, which can then be spatially manipulated using volumetric flow fields. We address the challenging question of how to aggregate information across different body poses, learning flow fields that allow for combining content from the appropriate regions of input images of highly non-rigid human subjects performing complex motions into a single implicit volumetric representation. This allows us to learn our 3D representation solely from videos of moving people. Armed with both 3D object understanding and end-to-end learned rendering, this categorically novel representation delivers state-of-the-art image generation quality, as shown by our quantitative and qualitative evaluations.

We propose novel motion representations for animating articulated objects consisting of distinct parts. In a completely unsupervised manner, our method identifies object parts, tracks them in a driving video, and infers their motions by considering their principal axes. In contrast to the previous keypoint-based works, our method extracts meaningful and consistent regions, describing locations, shape, and pose. The regions correspond to semantically relevant and distinct object parts, that are more easily detected in frames of the driving video. To force decoupling of foreground from background, we model non-object related global motion with an additional affine transformation. To facilitate animation and prevent the leakage of the shape of the driving object, we disentangle shape and pose of objects in the region space. Our model1can animate a variety of objects, surpassing previous methods by a large margin on existing benchmarks. We present a challenging new benchmark with high-resolution videos and show that the improvement is particularly pronounced when articulated objects are considered, reaching 96.6% user preference vs. the state of the art.

In deep model compression, the recent finding "Lottery Ticket Hypothesis" (LTH) pointed out that there could exist a winning ticket (i.e., a properly pruned sub-network together with original weight initialization) that can achieve competitive performance than the original dense network. However, it is not easy to observe such winning property in many scenarios, where for example, a relatively large learning rate is used even if it benefits training the original dense model. In this work, we investigate the underlying condition and rationale behind the winning property, and find that the underlying reason is largely attributed to the correlation between initialized weights and final-trained weights when the learning rate is not sufficiently large. Thus, the existence of winning property is correlated with an insufficient DNN pretraining, and is unlikely to occur for a well-trained DNN. To overcome this limitation, we propose the "pruning & fine-tuning" method that consistently outperforms lottery ticket sparse training under the same pruning algorithm and the same total training epochs. Extensive experiments over multiple deep models (VGG, ResNet, MobileNet-v2) on different datasets have been conducted to justify our proposals.

People usually take short videos to record meaningful moments in their lives. However, selecting the most representative frame, which not only has high image visual quality but also captures video content, from a short video to share or keep is a time-consuming process for one may need to manually go through all the frames in a video to make a decision. In this paper, we introduce the problem of the best frame selection in a short video and aim to solve it automatically. Towards this end, we collect and will release a diverse large-scale short video dataset that includes 11, 000 videos shoot in our daily life. All videos are assumed to be short (e.g., a few seconds) and each video has human-annotated of the best frame. Then we introduce a deep convolutional neural network (CNN) based approach with ranking objective to automatically pick the best frame from frame sequences extracted via short videos. Additionally, we propose new evaluation metrics, especially for the best frame selection. In experiments, we show our approach outperforms various other methods significantly.

Designing the structure of neural networks is considered one of the most challenging tasks in deep learning, especially when there is few prior knowledge about the task domain. In this paper, we propose an Ecologically-Inspired GENetic (EIGEN) approach that uses the concept of succession, extinction, mimicry, and gene duplication to search neural network structure from scratch with poorly initialized simple network and few constraints forced during the evolution, as we assume no prior knowledge about the task domain. Specifically, we first use primary succession to rapidly evolve a population of poorly initialized neural network structures into a more diverse population, followed by a secondary succession stage for fine-grained searching based on the networks from the primary succession. Extinction is applied in both stages to reduce computational cost. Mimicry is employed during the entire evolution process to help the inferior networks imitate the behavior of a superior network and gene duplication is utilized to duplicate the learned blocks of novel structures, both of which help to find better network structures. Experimental results show that our proposed approach can achieve similar or better performance compared to the existing genetic approaches with dramatically reduced computation cost. For example, the network discovered by our approach on CIFAR-100 dataset achieves 78.1% test accuracy under 120 GPU hours, compared to 77.0% test accuracy in more than 65, 536 GPU hours in [35].

Automatic image aesthetics rating has received a growing interest with the recent breakthrough in deep learning. Although many studies exist for learning a generic or universal aesthetics model, investigation of aesthetics models incorporating individual user's preference is quite limited. We address this personalized aesthetics problem by showing that individual's aesthetic preferences exhibit strong correlations with content and aesthetic attributes, and hence the deviation of individual's perception from generic image aesthetics is predictable. To accommodate our study, we first collect two distinct datasets, a large image dataset from Flickr and annotated by Amazon Mechanical Turk, and a small dataset of real personal albums rated by owners. We then propose a new approach to personalized aesthetics learning that can be trained even with a small set of annotated images from a user. The approach is based on a residual-based model adaptation scheme which learns an offset to compensate for the generic aesthetics score. Finally, we introduce an active learning algorithm to optimize personalized aesthetics prediction for real-world application scenarios. Experiments demonstrate that our approach can effectively learn personalized aesthetics preferences, and outperforms existing methods on quantitative comparisons.