Diffusion models have revolutionized the field of content synthesis and editing. Recent models have replaced the traditional UNet architecture with the Diffusion Transformer (DiT), and employed flow-matching for improved training and sampling. However, they exhibit limited generation diversity. In this work, we leverage this limitation to perform consistent image edits via selective injection of attention features. The main challenge is that, unlike the UNet-based models, DiT lacks a coarse-to-fine synthesis structure, making it unclear in which layers to perform the injection. Therefore, we propose an automatic method to identify "vital layers" within DiT, crucial for image formation, and demonstrate how these layers facilitate a range of controlled stable edits, from non-rigid modifications to object addition, using the same mechanism. Next, to enable real-image editing, we introduce an improved image inversion method for flow models. Finally, we evaluate our approach through qualitative and quantitative comparisons, along with a user study, and demonstrate its effectiveness across multiple applications.

Video personalization methods allow us to synthesize videos with specific concepts such as people, pets, and places. However, existing methods often focus on limited domains, require time-consuming optimization per subject, or support only a single subject. We present Video Alchemist—a video model with built-in multi-subject, openset personalization capabilities for both foreground objects and background, eliminating the need for time-consuming test-time optimization. Our model is built on a new Diffusion Transformer module that fuses each conditional reference image and its corresponding subject-level text prompt with cross-attention layers. Developing such a large model presents two main challenges: dataset and evaluation. First, as paired datasets of reference images and videos are extremely hard to collect, we sample selected video frames as reference images and synthesize a clip of the target video. However, while models can easily denoise training videos given reference frames, they fail to generalize to new contexts. To mitigate this issue, we design a new automatic data construction pipeline with extensive image augmentations. Second, evaluating open-set video personalization is a challenge in itself. To address this, we introduce a personalization benchmark that focuses on accurate subject fidelity and supports diverse personalization scenarios. Finally, our extensive experiments show that our method significantly outperforms existing personalization methods in both quantitative and qualitative evaluations.

We introduce Omni-ID, a novel facial representation designed specifically for generative tasks. Omni-ID encodes holistic information about an individual’s appearance across diverse expressions and poses within a fixed-size representation. It consolidates information from a varied number of unstructured input images into a structured representation, where each entry represents certain global or local identity features. Our approach uses a few-to-many identity reconstruction training paradigm, where a limited set of input images is used to reconstruct multiple target images of the same individual in various poses and expressions. A multi-decoder framework is introduced to leverage the complementary strengths of diverse decoders during training. Unlike conventional representations, such as ArcFace and CLIP, which are typically learned through discriminative or contrastive objectives, Omni-ID is optimized with a generative objective, resulting in a more comprehensive and nuanced identity capture for generative tasks. Trained on our MFHQ dataset – a multi-view facial image collection, Omni-ID demonstrates substantial improvements over conventional representations across various generative tasks.

This paper presents ThinkDiff, a novel alignment paradigm that empowers text-to-image diffusion models with multimodal in-context understanding and reasoning capabilities by integrating the strengths of vision-language models (VLMs). Existing multimodal diffusion finetuning methods largely focus on pixel-level reconstruction rather than in-context reasoning, and are constrained by the complexity and limited availability of reasoning-based datasets. ThinkDiff addresses these challenges by leveraging vision-language training as a proxy task, aligning VLMs with the decoder of an encoder-decoder large language model (LLM) instead of a diffusion decoder. This proxy task builds on the observation that the LLM decoder shares the same input feature space with diffusion decoders that use the corresponding LLM encoder for prompt embedding. As a result, aligning VLMs with diffusion decoders can be simplified through alignment with the LLM decoder. Without complex training and datasets, ThinkDiff effectively unleashes understanding, reasoning, and composing capabilities in diffusion models. Experiments demonstrate that ThinkDiff significantly improves accuracy from 19.2% to 46.3% on the challenging CoBSAT benchmark for multimodal in-context reasoning generation, with only 5 hours of training on 4 A100 GPUs. Additionally, ThinkDiff demonstrates exceptional performance in composing multiple images and texts into logically coherent images. Project page: https://mizhenxing.github.io/ThinkDiff.

Adapter-based training has emerged as a key mechanism for extending the capabilities of powerful foundation image generators, enabling personalized and stylized text-to-image synthesis. These adapters are typically trained to capture a specific target attribute, such as subject identity, using single-image reconstruction objectives. However, because the input image inevitably contains a mixture of visual factors, adapters are prone to entangle the target attribute with incidental ones, such as pose, expression, and lighting. This spurious correlation problem limits generalization and obstructs the model's ability to adhere to the input text prompt. In this work, we uncover a simple yet effective solution: provide the very shortcuts we wish to eliminate during adapter training. In Shortcut-Rerouted Adapter Training, confounding factors are routed through auxiliary modules, such as ControlNet or LoRA, eliminating the incentive for the adapter to internalize them. The auxiliary modules are then removed during inference. When applied to tasks like facial and full-body identity injection, our approach improves generation quality, diversity, and prompt adherence. These results point to a general design principle in the era of large models: when seeking disentangled representations, the most effective path may be to establish shortcuts for what should NOT be learned.

Recent advances in text-to-video generation have enabled high-quality synthesis from text and image prompts. While the personalization of dynamic concepts, which capture subject-specific appearance and motion from a single video, is now feasible, most existing methods require per-instance fine-tuning, limiting scalability. We introduce a fully zero-shot framework for dynamic concept personalization in text-to-video models. Our method leverages structured 2×2 video grids that spatially organize input and output pairs, enabling the training of lightweight Grid-LoRA adapters for editing and composition within these grids. At inference, a dedicated Grid Fill module completes partially observed layouts, producing temporally coherent and identity preserving outputs. Once trained, the entire system operates in a single forward pass, generalizing to previously unseen dynamic concepts without any test-time optimization. Extensive experiments demonstrate high-quality and consistent results across a wide range of subjects beyond trained concepts and editing scenarios.

We introduce a method for composing object-level visual prompts within a text-to-image diffusion model. Our approach addresses the task of generating semantically coherent compositions across diverse scenes and styles, similar to the versatility and expressiveness offered by text prompts. A key challenge in this task is to preserve the identity of the objects depicted in the input visual prompts, while also generating diverse compositions across different images. To address this challenge, we introduce a new KV-mixed cross-attention mechanism, in which keys and values are learned from distinct visual representations. The keys are derived from an encoder with a small bottleneck for layout control, whereas the values come from a larger bottleneck encoder that captures fine-grained appearance details. By mixing keys and values from these complementary sources, our model preserves the identity of the visual prompts while supporting flexible variations in object arrangement, pose, and composition. During inference, we further propose object-level compositional guidance to improve the method’s identity preservation and layout correctness. Results show that our technique produces diverse scene compositions that preserve the unique characteristics of each visual prompt, expanding the creative potential of text-to-image generation.

Generating high-fidelity images of humans with fine-grained control over attributes such as hairstyle and clothing remains a core challenge in personalized text-to-image synthesis. While prior methods emphasize identity preservation from a reference image, they lack modularity and fail to provide disentangled control over specific visual attributes. We introduce a new paradigm for attribute-specific image prompting, in which distinct sets of reference images are used to guide the generation of individual aspects of human appearance, such as hair, clothing, and identity. Our method encodes these inputs into attribute-specific tokens, which are injected into a pre-trained text-to-image diffusion model. This enables compositional and disentangled control over multiple visual factors, even across multiple people within a single image. To promote natural composition and robust disentanglement, we curate a cross-reference training dataset featuring subjects in diverse poses and expressions, and propose a multi-attribute cross-reference training strategy that encourages the model to generate faithful outputs from misaligned attribute inputs while adhering to both identity and textual conditioning. Extensive experiments show that our method achieves state-of-the-art performance in accurately following both visual and textual prompts. Our framework paves the way for more configurable human image synthesis by combining visual prompting with text-driven generation. Our project page is available at https://snap-research.github.io/composeme/.

Generative diffusion models can serve as priors, ensuring that image restoration solutions adhere to natural image manifolds. For facial images, however, personalized priors are essential to accurately reconstruct individual-specific facial features. We propose Dual-Pivot Tuning - a simple yet effective two-stage approach to personalize blind restoration systems while preserving general prior integrity. Our key observation is that for efficient personalization, the diffusion model should be tuned around a fixed textual pivot in the first step, while in the second step a guiding network should be tuned in a generic (non-personalized) manner, using the personalized diffusion model as a fixed "pivot". This approach ensures that personalization does not interfere with the restoration process, producing results with a natural appearance that show high fidelity to both identity and degraded image attributes. We conducted extensive experiments with images of widely recognized individuals, evaluating our approach both qualitatively and quantitatively against relevant baselines. Notably, our personalized prior not only achieves superior identity fidelity, but also outperforms state-of-the-art generic priors in terms of overall image quality. Project webpage is https://personalized-restoration.github.io/ and code is available at https://github.com/personalized-restoration/personalized-restoration.

We present 3D Paintbrush, a technique for automatically texturing local semantic regions on meshes via text descriptions. Our method is designed to operate directly on meshes, producing texture maps which seamlessly integrate into standard graphics pipelines. We opt to simultaneously produce a localization map (to specify the edit region) and a texture map which conforms to it. This approach improves the quality of both the localization and the stylization. To enhance the details and resolution of the textured area, we leverage multiple stages of a cascaded diffusion model to supervise our local editing technique with generative priors learned from images at different resolutions. Our technique, referred to as Cascaded Score Distillation (CSD), simultaneously distills scores at multiple resolutions in a cascaded fashion, enabling control over both the granularity and global understanding of the supervision. We demonstrate the effectiveness of 3D Paintbrush to locally texture different semantic regions on a variety of shapes. Project page: https://threedle.github.io/3d-paintbrush

Customization techniques for text-to-image models have paved the way for a wide range of previously unattainable applications, enabling the generation of specific concepts across diverse contexts and styles. While existing methods facilitate high-fidelity customization for individual concepts or a limited, pre-defined set of them, they fall short of achieving scalability, where a single model can seamlessly render countless concepts. In this paper, we address a new problem called Modular Customization, with the goal of efficiently merging customized models that were fine-tuned independently for individual concepts. This allows the merged model to Jointly synthesize concepts in one image without compromising fidelity or incurring any additional computational costs. To address this problem, we introduce Orthogonal Adaptation, a method designed to encourage the customized models, which do not have access to each other during fine-tuning, to have orthogonal residual weights. This ensures that during inference time, the customized models can be summed with minimal interference. Our proposed method is both simple and versatile, applicable to nearly all optimizable weights in the model architecture. Through an extensive set of quantitative and qualitative evaluations, our method consistently outperforms relevant baselines in terms of efficiency and identity preservation, demonstrating a significant leap toward scalable customization of diffusion models.

Personalization has emerged as a prominent aspect within the field of generative AI, enabling the synthesis of individuals in diverse contexts and styles, while retaining high-fidelity to their identities. However, the process of personalization presents inherent challenges in terms of time and memory requirements. Fine-tuning each personalized model needs considerable GPU time investment, and storing a personalized model per subject can be demanding in terms of storage capacity. To overcome these challenges, we propose HyperDreamBooth—a hypernetwork capable of efficiently generating a small set of personalized weights from a single image of a person. By composing these weights into the diffusion model, coupled with fast finetuning, HyperDreamBooth can generate a person's face in various contexts and styles, with high subject details while also preserving the model's crucial knowledge of diverse styles and semantic modifications. Our method achieves personalization on faces in roughly 20 seconds, 25x faster than DreamBooth and 125x faster than Textual Inversion, using as few as one reference image, with the same quality and style diversity as DreamBooth. Also our method yields a model that is 10,000x smaller than a normal DreamBooth model.

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.

We investigate the space of weights spanned by a large collection of customized diffusion models. We populate this space by creating a dataset of over 60,000 models, each of which is a base model fine-tuned to insert a different person's visual identity. We model the underlying manifold of these weights as a subspace, which we term $\textit{weights2weights}$. We demonstrate three immediate applications of this space that result in new diffusion models -- sampling, editing, and inversion. First, sampling a set of weights from this space results in a new model encoding a novel identity. Next, we find linear directions in this space corresponding to semantic edits of the identity (e.g., adding a beard), resulting in a new model with the original identity edited. Finally, we show that inverting a single image into this space encodes a realistic identity into a model, even if the input image is out of distribution (e.g., a painting). We further find that these linear properties of the diffusion model weight space extend to other visual concepts. Our results indicate that the weight space of fine-tuned diffusion models can behave as an interpretable $\textit{meta}$-latent space producing new models.

Abstract The field of visual computing is rapidly advancing due to the emergence of generative artificial intelligence (AI), which unlocks unprecedented capabilities for the generation, editing, and reconstruction of images, videos, and 3D scenes. In these domains, diffusion models are the generative AI architecture of choice. Within the last year alone, the literature on diffusion‐based tools and applications has seen exponential growth and relevant papers are published across the computer graphics, computer vision, and AI communities with new works appearing daily on arXiv. This rapid growth of the field makes it difficult to keep up with all recent developments. The goal of this state‐of‐the‐art report (STAR) is to introduce the basic mathematical concepts of diffusion models, implementation details and design choices of the popular Stable Diffusion model, as well as overview important aspects of these generative AI tools, including personalization, conditioning, inversion, among others. Moreover, we give a comprehensive overview of the rapidly growing literature on diffusion‐based generation and editing, categorized by the type of generated medium, including 2D images, videos, 3D objects, locomotion, and 4D scenes. Finally, we discuss available datasets, metrics, open challenges, and social implications. This STAR provides an intuitive starting point to explore this exciting topic for researchers, artists, and practitioners alike.

Recent advances in generative imagery have brought forth outpainting and inpainting models that can produce high-quality, plausible image content in unknown regions. However, the content these models hallucinate is necessarily inauthentic, since they are unaware of the true scene. In this work, we propose RealFill, a novel generative approach for image completion that fills in missing regions of an image with the content that should have been there. RealFill is a generative inpainting model that is personalized using only a few reference images of a scene. These reference images do not have to be aligned with the target image, and can be taken with drastically varying viewpoints, lighting conditions, camera apertures, or image styles. Once personalized, RealFill is able to complete a target image with visually compelling contents that are faithful to the original scene. We evaluate RealFill on a new image completion benchmark that covers a set of diverse and challenging scenarios, and find that it outperforms existing approaches by a large margin. Project page: https://realfill.github.io.

Recent large-scale text-guided diffusion models provide powerful image generation capabilities. Currently, a massive effort is given to enable the modification of these images using text only as means to offer intuitive and versatile editing tools. To edit a real image using these state-of-the-art tools, one must first invert the image with a meaningful text prompt into the pretrained model's domain. In this paper, we introduce an accurate inversion technique and thus facilitate an intuitive text-based modification of the image. Our proposed inversion consists of two key novel components: (i) Pivotal inversion for diffusion models. While current methods aim at mapping random noise samples to a single input image, we use a single pivotal noise vector for each timestamp and optimize around it. We demonstrate that a direct DDIM inversion is inadequate on its own, but does provide a rather good anchor for our optimization. (ii) Null-text optimization, where we only modify the unconditional textual embedding that is used for classifier-free guidance, rather than the input text embedding. This allows for keeping both the model weights and the conditional embedding intact and hence enables applying prompt-based editing while avoiding the cumbersome tuning of the model's weights. Our null-text inversion, based on the publicly available Stable Diffusion model, is extensively evaluated on a variety of images and various prompt editing, showing high-fidelity editing of real images.

The emergence of neural networks has revolutionized the field of motion synthesis. Yet, learning to unconditionally synthesize motions from a given distribution remains challenging, especially when the motions are highly diverse. In this work, we present MoDi - a generative model trained in an unsupervised setting from an extremely diverse, unstructured and unlabeled dataset. During inference, MoDi can synthesize high-quality, diverse motions. Despite the lack of any structure in the dataset, our model yields a well-behaved and highly structured latent space, which can be semantically clustered, constituting a strong motion prior that facilitates various applications including semantic editing and crowd animation. In addition, we present an encoder that inverts real motions into MoDi's natural motion manifold, issuing solutions to various ill-posed challenges such as completion from prefix and spatial editing. Our qualitative and quantitative experiments achieve state-of-the-art results that outperform recent SOTA techniques. Code and trained models are available at https://sigal-raab.github.io/MoDi.

Large text-to-image models achieved a remarkable leap in the evolution of AI, enabling high-quality and diverse synthesis of images from a given text prompt. However, these models lack the ability to mimic the appearance of subjects in a given reference set and synthesize novel renditions of them in different contexts. In this work, we present a new approach for “personalization” of text-to-image diffusion models. Given as input just a few images of a subject, we fine-tune a pretrained text-to-image model such that it learns to bind a unique identifier with that specific subject. Once the subject is embedded in the output domain of the model, the unique identifier can be used to synthesize novel photorealistic images of the subject contextualized in different scenes. By leveraging the semantic prior embedded in the model with a new autogenous class-specific prior preservation loss, our technique enables synthesizing the subject in diverse scenes, poses, views and lighting conditions that do not appear in the reference images. We apply our technique to several previously-unassailable tasks, including subject recontextualization, text-guided view synthesis, and artistic rendering, all while preserving the subject's key features. We also provide a new dataset and evaluation protocol for this new task of subject-driven generation. Project page: https://dreambooth.github.io/

This paper introduces Delta Denoising Score (DDS) a novel diffusion-based scoring technique which optimizes a parametric model for the task of image editing. Unlike the existing Score Distillation Sampling (SDS), which queries the generative model with a single image-text pair, DDS utilizes an additional fixed query of a reference image-text pair to generate delta scores that represent the difference between the outputs of the two queries. By estimating noisy gradient directions introduced by SDS using the source image and its text description, DDS provides cleaner gradient directions that modify the edited portions of the image while leaving others unchanged, thereby yielding a distilled edit of the source image. The analysis presented in this paper supports the power of the new score for image-to-image translation. We further show that the new score can be used to train an effective zero-shot image translation model. The experimental results show that the proposed loss term outperforms existing methods in terms of stability and quality, highlighting its potential for real-world applications.

We present DreamBooth3D, an approach to personalize text-to-3D generative models from as few as 3-6 casually captured images of a subject. Our approach combines recent advances in personalizing text-to-image models (DreamBooth) with text-to-3D generation (DreamFusion). We find that naïvely combining these methods fails to yield satisfactory subject-specific 3D assets due to personalized text-to-image models overfitting to the input viewpoints of the subject. We overcome this through a 3-stage optimization strategy where we jointly leverage the 3D consistency of neural radiance fields together with the personalization capability of text-to-image models. Our method can produce high-quality, subject-specific 3D assets with text-driven modifications such as novel poses, colors and attributes that are not seen in any of the input images of the subject. More results are available at our project page: https://dreambooth3d.github.io

Text-to-image model personalization aims to introduce a user-provided concept to the model, allowing its synthesis in diverse contexts. However, current methods primarily focus on the case of learning a single concept from multiple images with variations in backgrounds and poses, and struggle when adapted to a different scenario. In this work, we introduce the task of textual scene decomposition: given a single image of a scene that may contain several concepts, we aim to extract a distinct text token for each concept, enabling fine-grained control over the generated scenes. To this end, we propose augmenting the input image with masks that indicate the presence of target concepts. These masks can be provided by the user or generated automatically by a pre-trained segmentation model. We then present a novel two-phase customization process that optimizes a set of dedicated textual embeddings (handles), as well as the model weights, striking a delicate balance between accurately capturing the concepts and avoiding overfitting. We employ a masked diffusion loss to enable handles to generate their assigned concepts, complemented by a novel loss on cross-attention maps to prevent entanglement. We also introduce union-sampling, a training strategy aimed to improve the ability of combining multiple concepts in generated images. We use several automatic metrics to quantitatively compare our method against several baselines, and further affirm the results using a user study. Finally, we showcase several applications of our method.

Synthesizing human motion with a global structure, such as a choreography, is a challenging task. Existing methods tend to concentrate on local smooth pose transitions and neglect the global context or the theme of the motion. In this work, we present a music-driven motion synthesis framework that generates long-term sequences of human motions which are synchronized with the input beats, and jointly form a global structure that respects a specific dance genre. In addition, our framework enables generation of diverse motions that are controlled by the content of the music, and not only by the beat. Our music-driven dance synthesis framework is a hierarchical system that consists of three levels: pose, motif, and choreography. The pose level consists of an LSTM component that generates temporally coherent sequences of poses. The motif level guides sets of consecutive poses to form a movement that belongs to a specific distribution using a novel motion perceptual-loss. And the choreography level selects the order of the performed movements and drives the system to follow the global structure of a dance genre. Our results demonstrate the effectiveness of our music-driven framework to generate natural and consistent movements on various dance types, having control over the content of the synthesized motions, and respecting the overall structure of the dance.

Using only a model that was trained to predict where people look at images, and no additional training data, we can produce a range of powerful editing effects for reducing distraction in images. Given an image and a mask specifying the region to edit, we backpropagate through a state-of-the-art saliency model to parameterize a differentiable editing operator, such that the saliency within the masked region is reduced. We demonstrate several operators, including: a recoloring operator, which learns to apply a color transform that camouflages and blends distractors into their surroundings; a warping operator, which warps less salient image regions to cover distractors, gradually collapsing objects into themselves and effectively removing them (an effect akin to inpainting); a GAN operator, which uses a semantic prior to fully replace image regions with plausible, less salient alternatives. The resulting effects are consistent with cognitive research on the human visual system (e.g., since color mismatch is salient, the recoloring operator learns to harmonize objects' colors with their surrounding to reduce their saliency). And importantly, all effects are achieved under a zero-shot learning scenario, solely through the guidance of the pretrained saliency model, with no supervised data of the effects. We present results on a variety of natural images and conduct a perceptual study to evaluate and validate the changes in viewers' eye-gaze between the original images and our edited results. Project Webpage: https://deep-saliency-prior.github.io/

We present GANimator, a generative model that learns to synthesize novel motions from a single, short motion sequence. GANimator generates motions that resemble the core elements of the original motion, while simultaneously synthesizing novel and diverse movements. Existing data-driven techniques for motion synthesis require a large motion dataset which contains the desired and specific skeletal structure. By contrast, GANimator only requires training on a single motion sequence, enabling novel motion synthesis for a variety of skeletal structures e.g. , bipeds, quadropeds, hexapeds, and more. Our framework contains a series of generative and adversarial neural networks, each responsible for generating motions in a specific frame rate. The framework progressively learns to synthesize motion from random noise, enabling hierarchical control over the generated motion content across varying levels of detail. We show a number of applications, including crowd simulation, key-frame editing, style transfer, and interactive control, which all learn from a single input sequence. Code and data for this paper are at https://peizhuoli.github.io/ganimator.

We introduce MyStyle, a personalized deep generative prior trained with a few shots of an individual. MyStyle allows to reconstruct, enhance and edit images of a specific person, such that the output is faithful to the person's key facial characteristics. Given a small reference set of portrait images of a person (~ 100), we tune the weights of a pretrained StyleGAN face generator to form a local, low-dimensional, personalized manifold in the latent space. We show that this manifold constitutes a personalized region that spans latent codes associated with diverse portrait images of the individual. Moreover, we demonstrate that we obtain a personalized generative prior, and propose a unified approach to apply it to various ill-posed image enhancement problems, such as inpainting and super-resolution, as well as semantic editing. Using the personalized generative prior we obtain outputs that exhibit high-fidelity to the input images and are also faithful to the key facial characteristics of the individual in the reference set. We demonstrate our method with fair-use images of numerous widely recognizable individuals for whom we have the prior knowledge for a qualitative evaluation of the expected outcome. We evaluate our approach against few-shots baselines and show that our personalized prior, quantitatively and qualitatively, outperforms state-of-the-art alternatives.

Animating a newly designed character using motion capture (mocap) data is a long standing problem in computer animation. A key consideration is the skeletal structure that should correspond to the available mocap data, and the shape deformation in the joint regions, which often requires a tailored, pose-specific refinement. In this work, we develop a neural technique for articulating 3D characters using enveloping with a pre-defined skeletal structure which produces high quality pose dependent deformations. Our framework learns to rig and skin characters with the same articulation structure ( e.g. , bipeds or quadrupeds), and builds the desired skeleton hierarchy into the network architecture. Furthermore , we propose neural blend shapes - a set of corrective pose-dependent shapes which improve the deformation quality in the joint regions in order to address the notorious artifacts resulting from standard rigging and skinning. Our system estimates neural blend shapes for input meshes with arbitrary connectivity, as well as weighting coefficients which are conditioned on the input joint rotations. Unlike recent deep learning techniques which supervise the network with ground-truth rigging and skinning parameters, our approach does not assume that the training data has a specific underlying deformation model. Instead, during training, the network observes deformed shapes and learns to infer the corresponding rig, skin and blend shapes using indirect supervision. During inference, we demonstrate that our network generalizes to unseen characters with arbitrary mesh connectivity, including unrigged characters built by 3D artists. Conforming to standard skeletal animation models enables direct plug-and-play in standard animation software, as well as game engines.

We introduce MotioNet , a deep neural network that directly reconstructs the motion of a 3D human skeleton from a monocular video. While previous methods rely on either rigging or inverse kinematics (IK) to associate a consistent skeleton with temporally coherent joint rotations, our method is the first data-driven approach that directly outputs a kinematic skeleton, which is a complete, commonly used motion representation. At the crux of our approach lies a deep neural network with embedded kinematic priors, which decomposes sequences of 2D joint positions into two separate attributes: a single, symmetric skeleton encoded by bone lengths, and a sequence of 3D joint rotations associated with global root positions and foot contact labels. These attributes are fed into an integrated forward kinematics (FK) layer that outputs 3D positions, which are compared to a ground truth. In addition, an adversarial loss is applied to the velocities of the recovered rotations to ensure that they lie on the manifold of natural joint rotations. The key advantage of our approach is that it learns to infer natural joint rotations directly from the training data rather than assuming an underlying model, or inferring them from joint positions using a data-agnostic IK solver. We show that enforcing a single consistent skeleton along with temporally coherent joint rotations constrains the solution space, leading to a more robust handling of self-occlusions and depth ambiguities.

We introduce a novel deep learning framework for data-driven motion retargeting between skeletons, which may have different structure, yet corresponding to homeomorphic graphs. Importantly, our approach learns how to retarget without requiring any explicit pairing between the motions in the training set. We leverage the fact that different homeomorphic skeletons may be reduced to a common primal skeleton by a sequence of edge merging operations, which we refer to as skeletal pooling. Thus, our main technical contribution is the introduction of novel differentiable convolution, pooling, and unpooling operators. These operators are skeleton-aware , meaning that they explicitly account for the skeleton's hierarchical structure and joint adjacency, and together they serve to transform the original motion into a collection of deep temporal features associated with the joints of the primal skeleton. In other words, our operators form the building blocks of a new deep motion processing framework that embeds the motion into a common latent space, shared by a collection of homeomorphic skeletons. Thus, retargeting can be achieved simply by encoding to, and decoding from this latent space. Our experiments show the effectiveness of our framework for motion retargeting, as well as motion processing in general, compared to existing approaches. Our approach is also quantitatively evaluated on a synthetic dataset that contains pairs of motions applied to different skeletons. To the best of our knowledge, our method is the first to perform retargeting between skeletons with differently sampled kinematic chains, without any paired examples.

Transferring the motion style from one animation clip to another, while preserving the motion content of the latter, has been a long-standing problem in character animation. Most existing data-driven approaches are supervised and rely on paired data, where motions with the same content are performed in different styles. In addition, these approaches are limited to transfer of styles that were seen during training. In this paper, we present a novel data-driven framework for motion style transfer, which learns from an unpaired collection of motions with style labels, and enables transferring motion styles not observed during training. Furthermore, our framework is able to extract motion styles directly from videos, bypassing 3D reconstruction, and apply them to the 3D input motion. Our style transfer network encodes motions into two latent codes, for content and for style, each of which plays a different role in the decoding (synthesis) process. While the content code is decoded into the output motion by several temporal convolutional layers, the style code modifies deep features via temporally invariant adaptive instance normalization (AdaIN). Moreover, while the content code is encoded from 3D joint rotations, we learn a common embedding for style from either 3D or 2D joint positions, enabling style extraction from videos. Our results are comparable to the state-of-the-art, despite not requiring paired training data, and outperform other methods when transferring previously unseen styles. To our knowledge, we are the first to demonstrate style transfer directly from videos to 3D animations - an ability which enables one to extend the set of style examples far beyond motions captured by MoCap systems.

Abstract We present a new video‐based performance cloning technique. After training a deep generative network using a reference video capturing the appearance and dynamics of a target actor, we are able to generate videos where this actor reenacts other performances. All of the training data and the driving performances are provided as ordinary video segments, without motion capture or depth information. Our generative model is realized as a deep neural network with two branches, both of which train the same space‐time conditional generator, using shared weights. One branch, responsible for learning to generate the appearance of the target actor in various poses, uses paired training data, self‐generated from the reference video. The second branch uses unpaired data to improve generation of temporally coherent video renditions of unseen pose sequences. Through data augmentation, our network is able to synthesize images of the target actor in poses never captured by the reference video. We demonstrate a variety of promising results, where our method is able to generate temporally coherent videos, for challenging scenarios where the reference and driving videos consist of very different dance performances.

Analyzing human motion is a challenging task with a wide variety of applications in computer vision and in graphics. One such application, of particular importance in computer animation, is the retargeting of motion from one performer to another. While humans move in three dimensions, the vast majority of human motions are captured using video, requiring 2D-to-3D pose and camera recovery, before existing retargeting approaches may be applied. In this paper, we present a new method for retargeting video-captured motion between different human performers, without the need to explicitly reconstruct 3D poses and/or camera parameters. In order to achieve our goal, we learn to extract, directly from a video, a high-level latent motion representation, which is invariant to the skeleton geometry and the camera view. Our key idea is to train a deep neural network to decompose temporal sequences of 2D poses into three components: motion, skeleton, and camera view-angle. Having extracted such a representation, we are able to re-combine motion with novel skeletons and camera views, and decode a retargeted temporal sequence, which we compare to a ground truth from a synthetic dataset. We demonstrate that our framework can be used to robustly extract human motion from videos, bypassing 3D reconstruction, and outperforming existing retargeting methods, when applied to videos in-the-wild. It also enables additional applications, such as performance cloning, video-driven cartoons, and motion retrieval.

Correspondence between images is a fundamental problem in computer vision, with a variety of graphics applications. This paper presents a novel method for sparse cross-domain correspondence. Our method is designed for pairs of images where the main objects of interest may belong to different semantic categories and differ drastically in shape and appearance, yet still contain semantically related or geometrically similar parts. Our approach operates on hierarchies of deep features, extracted from the input images by a pre-trained CNN. Specifically, starting from the coarsest layer in both hierarchies, we search for Neural Best Buddies (NBB): pairs of neurons that are mutual nearest neighbors. The key idea is then to percolate NBBs through the hierarchy, while narrowing down the search regions at each level and retaining only NBBs with significant activations. Furthermore, in order to overcome differences in appearance, each pair of search regions is transformed into a common appearance. We evaluate our method via a user study, in addition to comparisons with alternative correspondence approaches. The usefulness of our method is demonstrated using a variety of graphics applications, including cross-domain image alignment, creation of hybrid images, automatic image morphing, and more.

Conventional synthetic aperture radar (SAR) systems are limited in their ability to satisfy the increasing requirement for improved spatial resolution and wider coverage. The demand for high resolution requires high sampling rates, while coverage is limited by the pulse repetition frequency. Consequently, sampling rate reduction is of high practical value in SAR imaging. In this paper, we introduce a new algorithm, equivalent to the well-known range-Doppler method, to process SAR data using the Fourier series coefficients of the raw signals. We then demonstrate how to exploit the algorithm features to reduce sampling rate in both range and azimuth axes and process the signals at sub-Nyquist rates, by using compressive sensing (CS) tools. In particular, we demonstrate recovery of an image using only a portion of the received signal's bandwidth and also while dropping a large percentage of the transmitted pulses. The complementary pulses may be used to capture other scenes within the same coherent processing interval. In addition, we propose exploiting the ability to reconstruct the image from narrow bands in order to dynamically adapt the transmitted waveform energy to vacant spectral bands, paving the way to cognitive SAR. The proposed recovery algorithms form a new CS-SAR imaging method that can be applied to high-resolution SAR data acquired at sub-Nyquist rates in range and azimuth. The performance of our method is assessed using simulated and real data sets. Finally, our approach is implemented in hardware using a previously suggested Xampling radar prototype.

The paper presents a novel three-dimensional shape acquisition and reconstruction method based on the well-known Archimedes equality between fluid displacement and the submerged volume. By repeatedly dipping a shape in liquid in different orientations and measuring its volume displacement, we generate the dip transform : a novel volumetric shape representation that characterizes the object's surface. The key feature of our method is that it employs fluid displacements as the shape sensor. Unlike optical sensors, the liquid has no line-of-sight requirements, it penetrates cavities and hidden parts of the object, as well as transparent and glossy materials, thus bypassing all visibility and optical limitations of conventional scanning devices. Our new scanning approach is implemented using a dipping robot arm and a bath of water, via which it measures the water elevation. We show results of reconstructing complex 3D shapes and evaluate the quality of the reconstruction with respect to the number of dips.