Multimodal Large Language Models struggle to maintain reliable performance under extreme real-world visual degradations, which impede their practical robustness. Existing robust MLLMs predominantly rely on implicit training/adaptation that focuses solely on visual encoder generalization, suffering from limited interpretability and isolated optimization. To overcome these limitations, we propose Robust-R1, a novel framework that explicitly models visual degradations through structured reasoning chains. Our approach integrates: (i) supervised fine-tuning for degradation-aware reasoning foundations, (ii) reward-driven alignment for accurately perceiving degradation parameters, and (iii) dynamic reasoning depth scaling adapted to degradation intensity. To facilitate this approach, we introduce a specialized 11K dataset featuring realistic degradations synthesized across four critical real-world visual processing stages, each annotated with structured chains connecting degradation parameters, perceptual influence, pristine semantic reasoning chain, and conclusion. Comprehensive evaluations demonstrate state-of-theart robustness: Robust-R1 outperforms all general and robust baselines on the real-world degradation benchmark R-Bench, while maintaining superior anti-degradation performance under multi-intensity adversarial degradations on MMMB, MMStar, and RealWorldQA.

Recent advances in multimodal LLMs and systems that use tools for long-video QA point to the promise of reasoning over hour-long episodes.However, many methods still compress content into lossy summaries or rely on limited toolsets, weakening temporal grounding and missing fine-grained cues.We propose a multi-agent framework in which a master LLM coordinates a grounding agent to localize question-relevant segments and a vision agent to extract targeted textual observations.The master agent plans with a step limit, and is trained with reinforcement learning to encourage concise, correct, and efficient multi-agent cooperation.This design helps the master agent focus on relevant clips via grounding, complements subtitles with visual detail, and yields interpretable trajectories.On our proposed LongTVQA and LongTVQA+ which are episode-level datasets aggregated from TVQA/TVQA+, our multiagent system significantly outperforms strong non-agent baselines.Experiments also show reinforcement learning further strengthens reasoning and planning for the trained agent.

Ke Ma, Jiaqi Tang, Bin Guo, Xueting Han, Ruonan Xu, Qingfeng He, Ziheng Wang, Xu Wang, Qifeng Chen, Zhiwen Yu, Yunhao Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Neural radiance fields, which represent a 3D scene as a color field and a density field, have demonstrated great progress in novel view synthesis yet are unfavorable for editing due to the implicitness. This work studies the task of efficient 3D editing, where we focus on editing speed and user interactivity. To this end, we propose to learn the color field as an explicit 2D appearance aggregation, also called canonical image, with which users can easily customize their 3D editing via 2D image processing. We complement the canonical image with a projection field that maps 3D points onto 2D pixels for texture query. This field is initialized with a pseudo canonical camera model and optimized with offset regularity to ensure the naturalness of the canonical image. Extensive experiments on different datasets suggest that our representation, dubbed AGAP, well supports various ways of 3D editing (e.g., stylization, instance segmentation, and interactive drawing). Our approach demonstrates remarkable efficiency by being at least 20× faster per edit compared to existing NeRF-based editing methods. Project page is available at h ttps: //felixcheng97.github.io/AGAP/.

This paper explores higher-resolution video outpainting with extensive content generation. We point out common issues faced by existing methods when attempting to largely outpaint videos: the generation of low-quality content and limitations imposed by GPU memory. To address these challenges, we propose a diffusion-based method called Infinite-Canvas. It builds upon two core designs. First, instead of employing the common practice of "single-shot" outpainting, we distribute the task across spatial windows and seamlessly merge them. It allows us to outpaint videos of any size and resolution without being constrained by GPU memory. Second, the source video and its relative positional relation are injected into the generation process of each window. It makes the generated spatial layout within each window harmonize with the source video. Coupling with these two designs enables us to generate higher-resolution outpainting videos with rich content while keeping spatial and temporal consistency. Infinite-Canvas excels in large-scale video outpainting, e.g., from 512 × 512 to 1152 × 2048 (9×), while producing high-quality and aesthetically pleasing results. It achieves the best quantitative results across various resolution and scale setups. The code is available at https://github.com/mayuelala/FollowYourCanvas.

Despite recent advances in UNet-based image editing, methods for shape-aware object editing in high-resolution images are still lacking. Compared to UNet, Diffusion Transformers (DiT) demonstrate superior capabilities to effectively capture the long-range dependencies among patches, leading to higher-quality image generation. In this paper, we propose DiT4Edit, the first Diffusion Transformer-based image editing framework. Specifically, DiT4Edit uses the DPM-Solver inversion algorithm to obtain the inverted latents, reducing the number of steps compared to the DDIM inversion algorithm commonly used in UNet-based frameworks. Additionally, we design unified attention control and patch merging, tailored for transformer computation streams. This integration allows our framework to generate higher-quality edited images faster. Our design leverages the advantages of DiT, enabling it to surpass UNet structures in image editing, especially in high-resolution and arbitrary-size images. Extensive experiments demonstrate the strong performance of DiT4Edit in various editing scenarios, highlighting the potential of diffusion transformers for image editing.

Despite recent advances in image-to-video generation, better controllability and local animation are less explored. Most existing image-to-video methods are not locally aware and tend to move the entire scene. However, human artists may need to control the movement of different objects or regions. Additionally, current I2V methods require users not only to describe the target motion but also to provide redundant detailed descriptions of frame contents.These two issues hinder the practical utilization of current I2V tools. In this paper, we propose a practical framework, named Follow-Your-Click, to achieve image animation with a simple user click (for specifying what to move) and a motion prompt (for specifying how to move). Technically, we propose the first-frame masking strategy, which significantly improves the video generation quality, and a motion-augmented module equipped with a motion prompt dataset to improve the motion prompt following abilities of our model. To further control the motion speed, we propose flow-based motion magnitude control to control the speed of target movement more precisely. Extensive experiments compared with 7 baselines, including both commercial tools and research methods on 8 metrics, suggest the superiority of our approach.

Due to the difficulty in scaling up, generative adversarial networks (GANs) seem to be falling out of grace with the task of text-conditioned image synthesis. Sparsely activated mixture-of-experts (MoE) has recently been demonstrated as a valid solution to training large-scale models with limited resources. Inspired by this, we present Aurora, a GAN-based text-to-image generator that employs a collection of experts to learn feature processing, together with a sparse router to adaptively select the most suitable expert for each feature point. We adopt a two-stage training strategy, which first learns a base model at 64 × 64 resolution followed by an upsampler to produce 512 × 512 images. Trained with only public data, our approach encouragingly closes the performance gap between GANs and industry-level diffusion models, maintaining a fast inference speed. We release the code and checkpoints here to facilitate the community for further development.

This work focuses on open-domain 4D avatarization, with the purpose of creating a 4D avatar from a portrait image in an arbitrary style. We select parametric triplanes as the intermediate 4D representation, and propose a practical training paradigm that takes advantage of both generative adversarial networks (GANs) and diffusion models. Our design stems from the observation that 4D GANs excel at bridging images and triplanes without supervision yet usually face challenges in handling diverse data distributions. A robust 2D diffusion prior emerges as the solution, assisting the GAN in transferring its expertise across various domains. The synergy between these experts permits the construction of a multi-domain image-triplane dataset, which drives the development of a general 4D avatar creator. Extensive experiments suggest that our model, termed AvatarArtist, is capable of producing high-quality 4D avatars with strong robustness to various source image domains. The code, the data, and the models will be made publicly available to facilitate future studies.

Derived from diffusion models, MangaNinja specializes in the task of reference-guided line art colorization. We incorporate two thoughtful designs to ensure precise character detail transcription, including a patch shuffling module to facilitate correspondence learning between the reference color image and the target line art, and a point-driven control scheme to enable fine-grained color matching. Experiments on a self-collected benchmark demonstrate the superiority of our model over current solutions in terms of precise colorization. We further showcase the potential of the proposed interactive point control in handling challenging cases (e.g., extreme poses and shadows), cross-character colorization, multi-reference harmonization, etc., beyond the reach of existing algorithms. Our code and model could be found here.

Recent progress in generative diffusion models has greatly advanced text-to-video generation. While text-to-video models trained on large-scale, diverse datasets can produce varied outputs, these generations often deviate from user preferences, highlighting the need for preference alignment on pre-trained models. Although Direct Preference Optimization (DPO) [42] has demonstrated significant improvements in language and image generation [52], we pioneer its adaptation to video diffusion models and propose a VideoDPO pipeline by making several key adjustments. Unlike previous image alignment methods that focus solely on either (i) visual quality or (ii) semantic alignment between text and videos, we comprehensively consider both dimensions and construct a preference score accordingly, which we term the OmniScore. We design a pipeline to automatically collect preference pair data based on the proposed OmniScore and discover that re-weighting these pairs based on the score significantly impacts overall preference alignment. Our experiments demonstrate substantial improvements in both visual quality and semantic alignment, ensuring that no preference aspect is neglected. Code and data are available at https://videodpo.github.io/.

As a highly practical application, image editing encounters a variety of user demands and thus prioritizes excellent ease of use. In this paper, we unveil MagicQuill, an integrated image editing system designed to support users in swiftly actualizing their creativity. Our system starts with a streamlined yet functionally robust interface, enabling users to articulate their ideas (e.g., inserting elements, erasing objects, altering color, etc.) with just a few strokes. These interactions are then monitored by a multimodal large language model (MLLM) to anticipate user intentions in real time, bypassing the need for prompt entry. Finally, we apply the powerful diffusion prior, enhanced by a carefully learned two-branch plug-in module, to process the editing request with precise control. Please visit the ${\text{project page}}$ to try out our system.

In this work, we systematically study music generation conditioned solely on the video. First, we present a large-scale dataset by collecting 360K video-music pairs, including various genres such as movie trailers, advertisements, and documentaries. Furthermore, we propose VidMuse, a simple framework for generating music aligned with video inputs. VidMuse stands out by producing high-fidelity music that is both acoustically and semantically aligned with the video. By incorporating local and global visual cues, VidMuse enables the creation of coherent music tracks that consistently match the video content through Long-Short-Term modeling. Through extensive experiments, VidMuse outperforms existing models in terms of audio quality, diversity, and audio-visual alignment. The code and datasets are available at https://vidmuse.github.io/

The intuitive nature of drag-based interaction has led to its growing adoption for controlling object trajectories in image-to-video synthesis. Still, existing methods that perform dragging in the 2D space usually face ambiguity when handling out-of-plane movements. In this work, we augment the interaction with a new dimension, i.e., the depth dimension, such that users are allowed to assign a relative depth for each point on the trajectory. That way, our new interaction paradigm not only inherits the convenience from 2D dragging, but facilitates trajectory control in the 3D space, broadening the scope of creativity. We propose a pioneering method for 3D trajectory control in image-to-video synthesis by abstracting object masks into a few cluster points. These points, accompanied by the depth information and the instance information, are finally fed into a video diffusion model as the control signal. Extensive experiments validate the effectiveness of our approach, dubbed LeviTor, in precisely manipulating the object movements when producing photo-realistic videos from static images. Our code is available at: https://github.com/ant-research/LeviTor.

Traditional reinforcement learning methods for human- object interaction (HOI) rely on labor-intensive, manually designed skill rewards that do not generalize well across different interactions. We introduce SkillMimic, a unified data-driven framework that fundamentally changes how agents learn interaction skills by eliminating the need for skill-specific rewards. Our key insight is that a unified HOI imitation reward can effectively capture the essence of diverse interaction patterns from HOI datasets. This enables SkillMimic to learn a single policy that not only masters multiple interaction skills but also facilitates skill transitions, with both diversity and generalization improving as the HOI dataset grows. For evaluation, we collect and introduce two basketball datasets containing approximately 35 minutes of diverse basketball skills. Extensive experiments show that SkillMimic successfully masters a wide range of basketball skills including stylistic variations in dribbling, layup, and shooting. Moreover, these learned skills can be effectively composed by a high-level controller to accomplish complex and long-horizon tasks such as consecutive scoring, opening new possibilities for scalable and generalizable interaction skill learning. Project page: https://ingrid789.github.io/SkillMimic/

As a verified need, consistent editing across in-the-wild images remains a technical challenge arising from various unmanageable factors, like object poses, lighting conditions, and photography environments. Edicho steps in with a training-free solution based on diffusion models, featuring a fundamental design principle of using explicit image correspondence to direct editing. Specifically, the key components include an attention manipulation module and a carefully refined classifier-free guidance (CFG) denoising strategy, both of which take into account the pre-estimated correspondence. Such an inference-time algorithm enjoys a plug-and-play nature and is compatible to most diffusion-based editing methods, such as ControlNet and BrushNet. Extensive results demonstrate the efficacy of Edicho in consistent cross-image editing under diverse settings. We will release the code to facilitate future studies.

Graphic design is crucial for conveying ideas and messages. Designers usually organize their work into objects, backgrounds, and vectorized text layers to simplify editing. However, this workflow demands considerable expertise. With the rise of GenAI methods, an endless supply of high-quality graphic designs in pixel format has become more accessible, though these designs often lack editability. Despite this, non-layered designs still inspire human designers, influencing their choices in layouts and text styles, ultimately guiding the creation of layered designs. Motivated by this observation, we propose Accordion, a graphic design generation framework taking the first attempt to convert AI-generated designs into editable layered designs, meanwhile refining nonsensical AI-generated text with meaningful alternatives guided by user prompts. It is built around a vision language model (VLM) playing distinct roles in three curated stages. For each stage, we design prompts to guide the VLM in executing different tasks. Distinct from existing bottom-up methods (e.g., COLE and Open-COLE) that gradually generate elements to create layered designs, our approach works in a top-down manner by using the visually harmonious reference image as global guidance to decompose each layer. Additionally, it leverages multiple vision experts such as SAM and element removal models to facilitate the creation of graphic layers. We train our method using the in-house graphic design dataset Design39K, augmented with AI-generated design images coupled with refined ground truth created by a customized inpainting model. Experimental results and user studies by designers show that Accordion generates favorable results on the DesignIntention benchmark, including tasks such as text-to-template, adding text to background, and text de-rendering, and also excels in creating design variations.

In light of the dynamic nature of autonomous driving environments and stringent safety requirements, general MLLMs combined with CLIP alone often struggle to accurately represent driving-specific scenarios, particularly in complex interactions and long-tail cases. To address this, we propose the Hints of Prompt (HoP) framework, which introduces three key enhancements: Affinity hint to emphasize instance-level structure by strengthening token-wise connections, Semantic hint to incorporate high-level information relevant to driving-specific cases, such as complex interactions among vehicles and traffic signs, and Question hint to align visual features with the query context, focusing on question-relevant regions. These hints are fused through a Hint Fusion module, enriching visual representations by capturing driving-related representations with limited domain data, ensuring faster adaptation to driving scenarios. Extensive experiments confirm the effectiveness of the HoP framework, showing that it significantly outperforms previous state-of-the-art methods in all key metrics.

Recent advances in Customized Concept Swapping (CCS) enable a text-to-image model to swap a concept in the source image with a customized target concept. However, the existing methods still face the challenges of $\textit{\textbf{inconsistency}}$ and $\textit{\textbf{inefficiency}}$. They struggle to maintain consistency in both the foreground and background during concept swapping, especially when the shape difference is large between objects. Additionally, they either require time-consuming training processes or involve redundant calculations during inference. To tackle these issues, we introduce InstantSwap, a new CCS method that aims to handle sharp shape disparity at speed. Specifically, we first extract the bbox of the object in the source image $\textit{automatically}$ based on attention map analysis and leverage the bbox to achieve both foreground and background consistency. For background consistency, we remove the gradient outside the bbox during the swapping process so that the background is free from being modified. For foreground consistency, we employ a cross-attention mechanism to inject semantic information into both source and target concepts inside the box. This helps learn semantic-enhanced representations that encourage the swapping process to focus on the foreground objects. To improve swapping speed, we avoid computing gradients at each timestep but instead calculate them periodically to reduce the number of forward passes, which improves efficiency a lot with a little sacrifice on performance. Finally, we establish a benchmark dataset to facilitate comprehensive evaluation. Extensive evaluations demonstrate the superiority and versatility of InstantSwap.

Enhancing a low-light noisy RAW image into a well-exposed and clean sRGB image is a significant challenge for modern digital cameras. Prior approaches have difficulties in recovering fine-grained details and true colors of the scene under extremely low-light environments due to near-to-zero SNR. Meanwhile, diffusion models have shown significant progress towards general domain image generation. In this paper, we propose to leverage the pre-trained latent diffusion model to perform the neural ISP for enhancing extremely low-light images. Specifically, to tailor the pre-trained latent diffusion model to operate on the RAW domain, we train a set of lightweight taming modules to inject the RAW information into the diffusion denoising process via modulating the intermediate features of UNet. We further observe different roles of UNet denoising and decoder reconstruction in the latent diffusion model, which inspires us to decompose the lowlight image enhancement task into latent-space low-frequency content generation and decoding-phase high-frequency detail maintenance. Through extensive experiments on representative datasets, we demonstrate our simple design not only achieves state-of-the-art performance in quantitative evaluations but also shows significant superiority in visual comparisons over strong baselines, which highlight the effectiveness of powerful generative priors for neural ISP under extremely low-light environments.

Tangram assembly, the art of human intelligence and manipulation dexterity, is a new challenge for robotics and reveals the limitations of state-of-the-arts. Here, we describe our initial exploration and highlight key problems in reasoning, planning, and manipulation for robotic tangram assembly. We present MRChaos (Master Rules from Chaos), a robust and general solution for learning assembly policies that can generalize to novel objects. In contrast to conventional methods based on prior geometric and kinematic models, MRChaos learns to assemble randomly generated objects through self-exploration in simulation without prior experience in assembling target objects. The reward signal is obtained from the visual observation change without manually designed models or annotations. MRChaos retains its robustness in assembling various novel tangram objects that have never been encountered during training, with only silhouette prompts. We show the potential of MRChaos in wider applications such as cutlery combinations. The presented work indicates that radical generalization in robotic assembly can be achieved by learning in much simpler domains. The code will be available https://robotll.github.io/MasterRulesFromChaos/.

Audio-driven video generation aims to synthesize realistic videos that align with input audio recordings, akin to the human ability to visualize scenes from auditory input. However, existing approaches predominantly focus on exploring semantic information, such as the classes of sounding sources present in the audio, limiting their ability to generate videos with accurate content and spatial composition. In contrast, we humans can not only naturally identify the semantic categories of sounding sources but also determine their deeply encoded spatial attributes, including locations and movement directions. This useful information can be elucidated by considering specific spatial indicators derived from the inherent physical properties of sound, such as loudness or frequency. As prior methods largely ignore this factor, we present SpA2V, the first framework explicitly exploits these spatial auditory cues from audios to generate videos with high semantic and spatial correspondence. SpA2V decomposes the generation process into two stages: 1) Audio-guided Video Planning: We meticulously adapt a state-of-the-art MLLM for a novel task of harnessing spatial and semantic cues from input audio to construct Video Scene Layouts (VSLs). This serves as an intermediate representation to bridge the gap between the audio and video modalities. 2) Layout-grounded Video Generation: We develop an efficient and effective approach to seamlessly integrate VSLs as conditional guidance into pre-trained diffusion models, enabling VSL-grounded video generation in a training-free manner. Extensive experiments demonstrate that SpA2V excels in generating realistic videos with semantic and spatial alignment to the input audios.

Recent advancements in video generation have enabled the creation of high-quality, visually compelling videos. However, generating videos that adhere to the laws of physics remains a critical challenge for applications requiring realism and accuracy. In this work, we propose **PhysHPO**, a novel framework for Hierarchical Cross-Modal Direct Preference Optimization, to tackle this challenge by enabling fine-grained preference alignment for physically plausible video generation. PhysHPO optimizes video alignment across four hierarchical granularities: a) ***Instance Level***, aligning the overall video content with the input prompt; b) ***State Level***, ensuring temporal consistency using boundary frames as anchors; c) ***Motion Level***, modeling motion trajectories for realistic dynamics; and d) ***Semantic Level***, maintaining logical consistency between narrative and visuals. Recognizing that real-world videos are the best reflections of physical phenomena, we further introduce an automated data selection pipeline to efficiently identify and utilize *"good data"* from existing large-scale text-video datasets, thereby eliminating the need for costly and time-intensive dataset construction. Extensive experiments on both physics-focused and general capability benchmarks demonstrate that PhysHPO significantly improves physical plausibility and overall video generation quality of advanced models. To the best of our knowledge, this is the first work to explore fine-grained preference alignment and data selection for video generation, paving the way for more realistic and human-preferred video generation paradigms.

Scroll images are a unique medium commonly used in virtual reality (VR) providing an immersive visual storytelling experience. Despite rapid advances in diffusion-based image generation, it remains an open research question to generate scroll images suitable for immersive, coherent, and controllable storytelling in VR. This paper proposes a multi-layered, diffusion-based scroll image generation framework with a novel semantic-aware denoising process. We incorporate layout prediction and style control modules to generate coherent scroll images of any aspect ratio. Based on the scroll image generation framework, we use different multi-window strategies to render diverse visual forms such as chains, rings, and forks for VR storytelling. Quantitative and qualitative evaluations demonstrate that our techniques can significantly enhance text-image consistency and visual coherence in scroll image generation, as well as the level of immersion and engagement of VR storytelling. We will release our source code to facilitate better collaborations on immersive storytelling between AI researchers and creative practitioners. https://magicscroll.github.io/

Text-based video editing has recently attracted considerable interest in changing the style or replacing the objects with a similar structure. Beyond this, we demonstrate that properties such as shape, size, location, motion, etc., can also be edited in videos. Our key insight is that the keyframe's transformations of the specific internal feature (e.g., edge maps of objects or human pose), can easily propagate to other frames to provide generation guidance. We thus propose MagicStick, a controllable video editing method that edits the video properties by utilizing the transformation on the extracted internal control signals. In detail, to keep the appearance, we inflate both the pre-trained image diffusion model and ControlNet to the temporal dimension and train low-rank adaptions (LoRA) layers to fit the specific scenes. Then, in editing, we perform an inversion and editing framework. Differently, finetuned ControlNet is introduced in both inversion and generation for attention guidance with the proposed attention remix between the spatial attention maps of inversion and editing. Yet succinct, our method is the first method to show the ability of video property editing from the pre-trained text-to-image model. We present experiments on numerous examples within our unified framework. We also compare with shape-aware text-based editing and handcrafted motion video generation, demonstrating our superior temporal consistency and editing capability than previous works.

High-quality portrait photography has become an essential function in our daily lives. However, due to the limited aperture and focal length of a smartphone camera, images captured by a smartphone cannot match the same level of bokeh effect by a digital single-lens reflex camera. A typical solution on a smartphone is to simulate out-of-focus effects from an all-in-focus image, where the key is robust depth estimation and portrait matting. To achieve this, we propose a multi-stage, multi-branch matting network to estimate a strand-level portrait alpha mask, which is then used to refine the coarse depth map obtained from the pre-trained model. Combining the input portrait image with the estimated depth map and alpha mask, we propose a learning-free optimization mechanism to construct a multi-plane image (MPI) representation for depth-of-field synthesis. The MPI consists of multiple layers of disk-blurred images with kernel size proportional to the absolute depth distance to the focus layer. Then a depth-aware blurring process is applied to enforce the bokeh effect. Besides, each MPI layer has an alpha channel controlling the visibility according to the corresponding depth. Finally, an image with bokeh is rendered by compositing all MPI layers. We conduct comprehensive experiments to evaluate our method, which demonstrates that our method can generate more accurate alpha masks and more realistic images with bokeh compared to prior work.

Abstract Considerable progress has been made in high dynamic range (HDR) image reconstruction from multi-exposure low dynamic range (LDR) frames in recent years. Despite achieving satisfactory HDR results while handling the misalignment among LDR frames, previous studies pay less attention to a prevalent but more crucial concern: the significant noise and motion blur that exist in multi-exposure frames captured by a handheld camera. To overcome these extensive and challenging corruptions, we achieve the HDR imaging task from two key aspects: First, due to the absence of related datasets, previous learning-based methods struggle with performing high-quality HDR imaging when the input LDR frames are confronted with real-world image noise and motion blur. Recognizing the importance of this aspect, we propose the first available realistic dataset based on the real-world burst imaging pipeline for training and evaluating different methods in the joint HDR imaging, denoising, and deblurring task. Second, due to the corruption-insensitive of previous network architectures, we propose a novel and efficient attention-based multi-exposure HDR imaging method, which skillfully selects the optimal information (clean or sharp) from corrupted LDR inputs by our customized cross-attention mechanism to generate HDR information. Furthermore, to enhance the robustness of our cross-attention mechanism, we introduce a novel E ntropy D ecreasing loss (ED loss) which decreases the entropy of the calculated attention map to alleviate the ghosting artifacts during multi-exposure information fusion. Extensive experimental results demonstrate that the proposed method trained on the proposed dataset surpasses related state-of-the-art methods with outstanding real-world photography quality. Codes and the dataset will be available.

Device-free localization (DFL) using easily-obtained Wi-Fi received signal strength (RSS) has wide real-world applications for not requiring people to carry trackable devices. However, accurate multitarget DFL remains challenging due to the unknown number of targets, multipath interference (MPI), especially between nearby targets, and limited real-world data. In this study, we pioneeringly propose a transformer-based learning method with Wi-Fi RSS as input, and an attentional prior fusion module, to simultaneously locate an unknown number of people at random positions. To overcome the multitarget data collection challenges, we contribute a large-scale cross-domain real-simulation-augmentation training dataset with one and two real-world nearby non-person objects at limited positions and up to five simulated and augmented randomly distributed targets. Experimental results demonstrate our method's improved accuracy, generalization ability, and robustness with fewer Wi-Fi nodes than previous methods.

Solving the task of inverse imaging problems can restore unknown clean images from input measurements that have incomplete information. Utilizing powerful generative models, such as denoising diffusion models, could better tackle the ill-posed issues of inverse problems with the distribution prior of the unknown clean images. We propose a learnable state-estimator-based diffusion model to incorporate the measurements into the reconstruction process. Our method makes efficient use of the pre-trained diffusion models with computational feasibility compared to the conditional diffusion models, which need to be trained from scratch. In addition, our pipeline does not require explicit knowledge of the image degradation operator or make the assumption of its form, unlike many other works that use the pre-trained diffusion models at the test time. The experiments on three typical inverse imaging problems (both linear and non-linear), inpainting, deblurring, and JPEG compression restoration, have comparable results with the state-of-the-art methods.

Generating text-editable and pose-controllable character videos have an imperious demand in creating various digital human. Nevertheless, this task has been restricted by the absence of a comprehensive dataset featuring paired video-pose captions and the generative prior models for videos. In this work, we design a novel two-stage training scheme that can utilize easily obtained datasets (i.e., image pose pair and pose-free video) and the pre-trained text-to-image (T2I) model to obtain the pose-controllable character videos. Specifically, in the first stage, only the keypoint image pairs are used only for a controllable text-to-image generation. We learn a zero-initialized convolutional encoder to encode the pose information. In the second stage, we finetune the motion of the above network via a pose-free video dataset by adding the learnable temporal self-attention and reformed cross-frame self-attention blocks. Powered by our new designs, our method successfully generates continuously pose-controllable character videos while keeps the editing and concept composition ability of the pre-trained T2I model. The code and models are available on https://follow-your-pose.github.io/.

Efficient generation of 3D digital humans is important in several industries, including virtual reality, social media, and cinematic production. 3D generative adversarial net-works (GANs) have demonstrated state-of-the-art (SOTA) quality and diversity for generated assets. Current 3D GAN architectures, however, typically rely on volume representations, which are slow to render, thereby hampering the GAN training and requiring multi- view-inconsistent 2D upsam-plers. Here, we introduce Gaussian Shell Maps (GSMs) as a framework that connects SOTA generator network archi-tectures with emerging 3D Gaussian rendering primitives using an articulable multi shell-based scaffold. In this set-ting, a CNN generates a 3D texture stack with features that are mapped to the shells. The latter represent inflated and deflated versions of a template surface of a digital human in a canonical body pose. Instead of rasterizing the shells directly, we sample 3D Gaussians on the shells whose at-tributes are encoded in the texture features. These Gaus-sians are efficiently and differentiably rendered. The ability to articulate the shells is important during GAN training and, at inference time, to deform a body into arbitrary user-defined poses. Our efficient rendering scheme bypasses the need for view-inconsistent upsamplers and achieves high-quality multi-view consistent renderings at a native resolution of 512 ×512 pixels. We demonstrate that GSMs suc-cessfully generate 3D humans when trained on single-view datasets, including SHHQ and DeepFashion. Project Page: rameenabdal.github.io/GaussianShellMaps

We propose DiffSHEG, a Diffusion-based approach for Speech-driven Holistic 3D Expression and Gesture generation with arbitrary length. While previous works focused on co-speech gesture or expression generation individually, the joint generation of synchronized expressions and gestures remains barely explored. To address this, our diffusion-based co-speech motion generation transformer enables uni-directional information flow from expression to gesture, facilitating improved matching of joint expression-gesture distributions. Furthermore, we introduce an outpainting-based sampling strategy for arbitrary long sequence generation in diffusion models, offering flexibility and computational efficiency. Our method provides a practical solution that produces high-quality synchronized expression and gesture generation driven by speech. Evaluated on two public datasets, our approach achieves state-of-the-art performance both quantitatively and qualitatively. Additionally, a user study confirms the superiority of DiffSHEG over prior approaches. By enabling the real-time generation of expressive and synchronized motions, DiffSHEG showcases its potential for various applications in the development of digital humans and embodied agents.

We propose a framework for automatic colorization that allows for iterative editing and modifications. The core of our framework lies in an imagination module: by understanding the content within a grayscale image, we utilize a pre-trained image generation model to generate multiple images that contain the same content. These images serve as references for coloring, mimicking the process of human experts. As the synthesized images can be imperfect or different from the original grayscale image, we propose a Reference Refinement Module to select the optimal reference composition. Unlike most previous end-to-end automatic colorization algorithms, our framework allows for iterative and localized modifications of the colorization results because we explicitly model the coloring samples. Extensive experiments demonstrate the superiority of our framework over existing automatic colorization algorithms in editability and flexibility. Project page: https://xy-cong.github.io/imagine-colorization/.

Existing depth sensors are imperfect and may provide inaccurate depth values in challenging scenarios, such as in the presence of transparent or reflective objects. In this work, we present a general framework that leverages polarization imaging to improve inaccurate depth measurements from various depth sensors. Previous polarization-based depth enhancement methods focus on utilizing pure physics-based formulas for a single sensor. In contrast, our method first adopts a learning-based strategy where a neural network is trained to estimate a dense and complete depth map from polarization data and a sensor depth map from different sensors. To further improve the performance, we propose a Polarization Prompt Fusion Tuning (PPFT) strategy to effectively utilize RGB-based models pre-trained on large-scale datasets, as the size of the polarization dataset is limited to train a strong model from scratch. We conducted extensive experiments on a public dataset, and the results demonstrate that the proposed method performs favorably compared to existing depth enhancement baselines. Code and demos are available at https://lastbasket.github.io/PPFT/.

We present the content deformation field (CoDeF) as a new type of video representation, which consists of a canonical content field aggregating the static contents in the entire video and a temporal deformation field recording the transformations from the canonical image (i.e., rendered from the canonical content field) to each individual frame along the time axis. Given a target video, these two fields are jointly optimized to reconstruct it through a carefully tailored rendering pipeline. We advisedly introduce some regularizations into the optimization process, urging the canonical content field to inherit semantics (e.g., the object shape) from the video. With such a design, CoDeF naturally supports lifting image algorithms for video processing, in the sense that one can apply an image algorithm to the canonical image and effortlessly propagate the outcomes to the entire video with the aid of the temporal deformation field. We experimentally show that CoDeF is able to lift image-to-image translation to video-to-video translation and lift keypoint detection to keypoint tracking without any training. More importantly, thanks to our lifting strategy that deploys the algorithms on only one image, we achieve superior cross-frame consistency in processed videos compared to existing video-to-video translation approaches, and even manage to track non-rigid objects like water and smog. Code is made available at https: / /qiuyu96. github.io/CoDeF/

Video and audio content creation serves as the core technique for the movie industry and professional users. Re-cently, existing diffusion-based methods tackle video and audio generation separately, which hinders the technique transfer from academia to industry. In this work, we aim at filling the gap, with a carefully designed optimization-based framework for cross-visual-audio and joint-visual-audio generation. We observe the powerful generation abil-ity of off-the-shelf video or audio generation models. Thus, instead of training the giant models from scratch, we pro-pose to bridge the existing strong models with a shared la-tent representation space. Specifically, we propose a mul-timodality latent aligner with the pre-trained ImageBind model. Our latent aligner shares a similar core as the clas-sifier guidance that guides the diffusion denoising process during inference time. Through carefully designed opti-mization strategy and loss functions, we show the superior performance of our method on joint video-audio generation, visual-steered audio generation, and audio-steered vi-sual generation tasks. The project website can be found at https://yzxing87.github.io/Seeing-and-Hearing/.

In this work, we investigate the capability of generating images from pre-trained diffusion models at much higher resolutions than the training image sizes. In addition, the generated images should have arbitrary image aspect ratios. When generating images directly at a higher resolution, 1024 x 1024, with the pre-trained Stable Diffusion using training images of resolution 512 x 512, we observe persistent problems of object repetition and unreasonable object structures. Existing works for higher-resolution generation, such as attention-based and joint-diffusion approaches, cannot well address these issues. As a new perspective, we examine the structural components of the U-Net in diffusion models and identify the crucial cause as the limited perception field of convolutional kernels. Based on this key observation, we propose a simple yet effective re-dilation that can dynamically adjust the convolutional perception field during inference. We further propose the dispersed convolution and noise-damped classifier-free guidance, which can enable ultra-high-resolution image generation (e.g., 4096 x 4096). Notably, our approach does not require any training or optimization. Extensive experiments demonstrate that our approach can address the repetition issue well and achieve state-of-the-art performance on higher-resolution image synthesis, especially in texture details. Our work also suggests that a pre-trained diffusion model trained on low-resolution images can be directly used for high-resolution visual generation without further tuning, which may provide insights for future research on ultra-high-resolution image and video synthesis. More results are available at the anonymous website: https://scalecrafter.github.io/ScaleCrafter/

Large Language Models (LLMs) and Large Multi-modality Models (LMMs) have demonstrated remarkable decision masking capabilities on a variety of tasks. However, they inherently operate planning within the language space, lacking the vision and spatial imagination ability. In contrast, humans utilize both left and right hemispheres of the brain for language and visual planning during the thinking process. Therefore, we introduce a novel vision-language planning framework in this work to perform concurrent visual and language planning for tasks with inputs of any form. Our framework incorporates visual planning to capture intricate environmental details, while language planning enhances the logical coherence of the overall system. We evaluate the effectiveness of our framework across vision-language tasks, vision-only tasks, and language-only tasks. The results demonstrate the superior performance of our approach, indicating that the integration of visual and language planning yields better contextually aware task execution.

We present a new 3D point-based detector model, named Shift-SSD, for precise 3D object detection in autonomous driving. Traditional point-based 3D object detectors often employ architectures that rely on a progressive downsampling of points. While this method effectively reduces computational demands and increases receptive fields, it will compromise the preservation of crucial non-local information for accurate 3D object detection, especially in the complex driving scenarios. To address this, we introduce an intriguing Cross-Cluster Shifting operation to unleash the representation capacity of the point-based detector by efficiently modeling longer-range inter-dependency while including only a negligible overhead. Concretely, the Cross-Cluster Shifting operation enhances the conventional design by shifting partial channels from neighboring clusters, which enables richer interaction with non-local regions and thus enlarges the receptive field of clusters. We conduct extensive experiments on the KITTI, Waymo, and nuScenes datasets, and the results demonstrate the state-of-the-art performance of Shift-SSD in both detection accuracy and runtime efficiency.

We present TALE, a novel training-free framework harnessing the generative capabilities of text-to-image diffusion models to address the cross-domain image composition task that focuses on flawlessly incorporating user-specified objects into a designated visual

Video Anomaly Detection (VAD) systems can autonomously monitor and identify disturbances, reducing the need for manual labor and associated costs. However, current VAD systems are often limited by their superficial semantic understanding of scenes and minimal user interaction. Additionally, the prevalent data scarcity in existing datasets restricts their applicability in open-world scenarios. In this paper, we introduce HAWK, a novel framework that leverages interactive large Visual Language Models (VLM) to interpret video anomalies precisely. Recognizing the difference in motion information between abnormal and normal videos, HAWK explicitly integrates motion modality to enhance anomaly identification. To reinforce motion attention, we construct an auxiliary consistency loss within the motion and video space, guiding the video branch to focus on the motion modality. Moreover, to improve the interpretation of motion-to-language, we establish a clear supervisory relationship between motion and its linguistic representation. Furthermore, we have annotated over 8,000 anomaly videos with language descriptions, enabling effective training across diverse open-world scenarios, and also created 8,000 question-answering pairs for users' open-world questions. The final results demonstrate that HAWK achieves SOTA performance, surpassing existing baselines in both video description generation and question-answering. Our codes/dataset/demo will be released at https://github.com/jqtangust/hawk.

This paper studies the problem of estimating physical properties (system identification) through visual observations. To facilitate geometry-aware guidance in physical property estimation, we introduce a novel hybrid framework that leverages 3D Gaussian representation to not only capture explicit shapes but also enable the simulated continuum to render object masks as 2D shape surrogates during training. We propose a new dynamic 3D Gaussian framework based on motion factorization to recover the object as 3D Gaussian point sets across different time states. Furthermore, we develop a coarse-to-fine filling strategy to generate the density fields of the object from the Gaussian reconstruction, allowing for the extraction of object continuums along with their surfaces and the integration of Gaussian attributes into these continuum. In addition to the extracted object surfaces, the Gaussian-informed continuum also enables the rendering of object masks during simulations, serving as 2D-shape guidance for physical property estimation. Extensive experimental evaluations demonstrate that our pipeline achieves state-of-the-art performance across multiple benchmarks and metrics. Additionally, we illustrate the effectiveness of the proposed method through real-world demonstrations, showcasing its practical utility. Our project page is at https://jukgei.github.io/project/gic.

Different camera sensors have different noise patterns, and thus an image denoising model trained on one sensor often does not generalize well to a different sensor. One plausible solution is to collect a large dataset for each sensor for training or fine-tuning, which is inevitably time-consuming. To address this cross-domain challenge, we present a novel adaptive domain learning (ADL) scheme for cross-domain RAW image denoising by utilizing existing data from different sensors (source domain) plus a small amount of data from the new sensor (target domain). The ADL training scheme automatically removes the data in the source domain that are harmful to fine-tuning a model for the target domain (some data are harmful as adding them during training lowers the performance due to domain gaps). Also, we introduce a modulation module to adopt sensor-specific information (sensor type and ISO) to understand input data for image denoising. We conduct extensive experiments on public datasets with various smartphone and DSLR cameras, which show our proposed model outperforms prior work on cross-domain image denoising, given a small amount of image data from the target domain sensor.

Humanoid robots are promising approach to automating patient interviews routinely conducted by medical staff. Their human-like appearance enables them to use the full gamut of verbal and behavioral cues that are critical to a successful interview. On the other hand, anthropomorphism can induce expectations of human-level performance by the robot. Not meeting such expectations degrades the quality of interaction. Specifically, humans expect rich real-time interactions during speech exchange, such as backchanneling and barge-ins. The nature of the patient interview task differs from most other scenarios where task oriented dialogue systems have been used, as there is increased potential of engagement breakdown during interaction. We describe a dialogue system architecture that improves the performance of humanoid robots on the patient interview task. Our architecture adds a nested inner real-time control loop to improve the timeliness of the robot’s responses based on the notion of "stance", an elaboration of the concept of a "turn", common in most existing dialogue systems. It also expands the dialogue state to monitor not only task progress, but also human engagement. Experiments using a humanoid robot running our proposed architecture reveal improved performance on interview tasks in terms of the perceived timeliness of responses and users’ impressions of the system.

We present Follow-Your-Emoji, a diffusion-based framework for portrait animation, which animates a reference portrait with target landmark sequences. The main challenge of portrait animation is to preserve the identity of the reference portrait and transfer the target expression to this portrait while maintaining temporal consistency and fidelity. To address these challenges, Follow-Your-Emoji equipped the powerful Stable Diffusion model with two well-designed technologies. Specifically, we first adopt a new explicit motion signal, namely expression-aware landmark, to guide the animation process. We discover this landmark can not only ensure the accurate motion alignment between the reference portrait and target motion during inference but also increase the ability to portray exaggerated expressions (i.e., large pupil movements) and avoid identity leakage. Then, we propose a facial fine-grained loss to improve the model’s ability of subtle expression perception and reference portrait appearance reconstruction by using both expression and facial masks. Accordingly, our method demonstrates significant performance in controlling the expression of freestyle portraits, including real humans, cartoons, sculptures, and even animals. By leveraging a simple and effective progressive generation strategy, we extend our model to stable long-term animation, thus increasing its potential application value. To address the lack of a benchmark for this field, we introduce EmojiBench, a comprehensive benchmark comprising diverse portrait images, driving videos, and landmarks. We show extensive evaluations on EmojiBench to verify the superiority of Follow-Your-Emoji. The code, training dataset and benchmark will be found in https://github.com/mayuelala/FollowYourEmoji.

We present ENTED, a new framework for blind face restoration that aims to restore high-quality and realistic portrait images. Our method involves repairing a single degraded input image using a high-quality reference image. We utilize a texture extraction and distribution framework to transfer high-quality texture features between the degraded input and reference image. However, the StyleGAN-like architecture in our framework requires high-quality latent codes to generate realistic images. The latent code extracted from the degraded input image often contains corrupted features, making it difficult to align the semantic information from the input with the high-quality textures from the reference. To overcome this challenge, we employ two special techniques. The first technique, inspired by vector quantization, replaces corrupted semantic features with high-quality code words. The second technique generates style codes that carry photorealistic texture information from a more informative latent space developed using the high-quality features in the reference image’s manifold. Extensive experiments conducted on synthetic and real-world datasets demonstrate that our method produces results with more realistic contextual details and outperforms state-of-the-art methods. A thorough ablation study confirms the effectiveness of each proposed module.

Generative Adversarial Networks (GANs) have significantly advanced image synthesis through mapping randomly sampled latent codes to high-fidelity synthesized images. However, applying well-trained GANs to real image editing remains challenging. A common solution is to find an approximate latent code that can adequately recover the input image to edit, which is also known as GAN inversion. To invert a GAN model, prior works typically focus on reconstructing the target image at the pixel level, yet few studies are conducted on whether the inverted result can well support manipulation at the semantic level. This work fills in this gap by proposing in-domain GAN inversion, which consists of a domain-guided encoder and a domain-regularized optimizer, to regularize the inverted code in the native latent space of the pre-trained GAN model. In this way, we manage to sufficiently reuse the knowledge learned by GANs for image reconstruction, facilitating a wide range of editing applications without any retraining. We further make comprehensive analyses on the effects of the encoder structure, the starting inversion point, as well as the inversion parameter space, and observe the trade-off between the reconstruction quality and the editing property. Such a trade-off sheds light on how a GAN model represents an image with various semantics encoded in the learned latent distribution.

Open-set action recognition is to reject unknown human action cases which are out of the distribution of the training set. Existing methods mainly focus on learning better uncertainty scores but dismiss the importance of feature representations. We find that features with richer semantic diversity can significantly improve the open-set performance under the same uncertainty scores. In this paper, we begin with analyzing the feature representation behavior in the open-set action recognition (OSAR) problem based on the information bottleneck (IB) theory, and propose to enlarge the instance-specific (IS) and classspecific (CS) information contained in the feature for better performance. To this end, a novel Prototypical Similarity Learning (PSL) framework is proposed to keep the instance variance within the same class to retain more IS information. Besides, we notice that unknown samples sharing similar appearances to known samples are easily misclassified as known classes. To alleviate this issue, video shuffling is further introduced in our PSL to learn distinct temporal information between original and shuffled samples, which we find enlarges the CS information. Extensive experiments demonstrate that the proposed PSL can significantly boost both the open-set and closed-set performance and achieves state-of-the-art results on multiple benchmarks. Code is available at https://github.com/Jun-CEN/PSL.

Many videos contain flickering artifacts; common causes of flicker include video processing algorithms, video generation algorithms, and capturing videos under specific situations. Prior work usually requires specific guidance such as the flickering frequency, manual annotations, or extra consistent videos to remove the flicker. In this work, we propose a general flicker removal framework that only receives a single flickering video as input without additional guidance. Since it is blind to a specific flickering type or guidance, we name this “blind deflickering.” The core of our approach is utilizing the neural atlas in cooperation with a neural filtering strategy. The neural atlas is a unified representation for all frames in a video that provides temporal consistency guidance but is flawed in many cases. To this end, a neural network is trained to mimic a filter to learn the consistent features (e.g., color, brightness) and avoid introducing the artifacts in the atlas. To validate our method, we construct a dataset that contains diverse real-world flickering videos. Extensive experiments show that our method achieves satisfying deflickering performance and even outperforms baselines that use extra guidance on a public benchmark. The source code is publicly available at https://chenyanglei.github.io/deflicker.

GAN inversion and editing via StyleGAN maps an input image into the embedding spaces (W, W+, and F) to simultaneously maintain image fidelity and meaningful manipulation. From latent space W to extended latent space W+ to feature space F in StyleGAN, the editability of GAN inversion decreases while its reconstruction quality increases. Recent GAN inversion methods typically explore W+ and F rather than W to improve reconstruction fidelity while maintaining editability. As W+ and F are derived from W that is essentially the foundation latent space of StyleGAN, these GAN inversion methods focusing on W+ and F spaces could be improved by stepping back to W. In this work, we propose to first obtain the proper latent code in foundation latent space W. We introduce contrastive learning to align W and the image space for proper latent code discovery. Then, we leverage a cross-attention encoder to transform the obtained latent code in W into W+ and F, accordingly. Our experiments show that our exploration of the foundation latent space W improves the representation ability of latent codes in W+ and features in F, which yields state-of-the-art reconstruction fidelity and editability results on the standard benchmarks. Project page: https://kumapowerliu.github.io/CLCAE.

The Federated Learning (FL) paradigm is known to face challenges under heterogeneous client data. Local training on non-iid distributed data results in deflected local optimum, which causes the client models drift further away from each other and degrades the aggregated global model's performance. A natural solution is to gather all client data onto the server, such that the server has a global view of the entire data distribution. Unfortunately, this reduces to regular training, which compromises clients' privacy and conflicts with the purpose of FL. In this paper, we put forth an idea to collect and leverage global knowledge on the server without hindering data privacy. We unearth such knowledge from the dynamics of the global model's trajectory. Specifically, we first reserve a short trajectory of global model snapshots on the server. Then, we synthesize a small pseudo dataset such that the model trained on it mimics the dynamics of the reserved global model trajectory. Afterward, the synthesized data is used to help aggregate the deflected clients into the global model. We name our method Dynafed, which enjoys the following advantages: 1) we do not rely on any external on-server dataset, which requires no additional cost for data collection; 2) the pseudo data can be synthesized in early communication rounds, which enables Dyna fedto take effect early for boosting the convergence and stabilizing training; 3) the pseudo data only needs to be synthesized once and can be directly utilized on the server to help aggregation in subsequent rounds. Experiments across extensive benchmarks are conducted to showcase the effectiveness of Dyna fed. We also provide insights and understanding of the underlying mechanism of our method.

Contemporary image rescaling aims at embedding a high-resolution (HR) image into a low-resolution (LR) thumbnail image that contains embedded information for HR image reconstruction. Unlike traditional image super-resolution, this enables high-fidelity HR image restoration faithful to the original one, given the embedded information in the LR thumbnail. However, state-of-the-art image rescaling methods do not optimize the LR image file size for efficient sharing and fall short of real-time performance for ultra-high-resolution (e.g., 6K) image reconstruction. To address these two challenges, we propose a novel frame-work (HyperThumbnail) for real-time 6K rate-distortion-aware image rescaling. Our framework first embeds an HR image into a JPEG LR thumbnail by an encoder with our proposed quantization prediction module, which minimizes the file size of the embedding LR JPEG thumbnail while maximizing HR reconstruction quality. Then, an efficient frequency-aware decoder reconstructs a high-fidelity HR image from the LR one in real time. Extensive experiments demonstrate that our framework outperforms previous image rescaling baselines in rate-distortion performance and can perform 6K image reconstruction in real time.

Existing methods for 3D-aware image synthesis largely depend on the 3D pose distribution pre-estimated on the training set. An inaccurate estimation may mislead the model into learning faulty geometry. This work proposes PoF3D that frees generative radiance fields from the requirements of 3D pose priors. We first equip the generator with an efficient pose learner, which is able to infer a pose from a latent code, to approximate the underlying true pose distribution automatically. We then assign the discriminator a task to learn pose distribution under the supervision of the generator and to differentiate real and synthesized images with the predicted pose as the condition. The pose-free generator and the pose-aware discriminator are jointly trained in an adversarial manner. Extensive results on a couple of datasets confirm that the performance of our approach, regarding both image quality and geometry quality, is on par with state of the art. To our best knowledge, PoF3D demonstrates the feasibility of learning high-quality 3D-aware image synthesis without using 3D pose priors for the first time. Project page can be found here.

This paper presents a 3D diffusion model that automatically generates 3D digital avatars represented as neural radiance fields (NeRFs). A significant challenge for 3D diffusion is that the memory and processing costs are prohibitive for producing high-quality results with rich details. To tackle this problem, we propose the roll-out diffusion network (RODIN), which takes a 3D NeRF model represented as multiple 2D feature maps and rolls out them onto a single 2D feature plane within which we perform 3D-aware diffusion. The RODIN model brings much-needed computational efficiency while preserving the integrity of 3D diffusion by using 3D-aware convolution that attends to projected features in the 2D plane according to their original relationships in 3D. We also use latent conditioning to orchestrate the feature generation with global coherence, leading to high-fidelity avatars and enabling semantic editing based on text prompts. Finally, we use hierarchical synthesis to further enhance details. The 3D avatars generated by our model compare favorably with those produced by existing techniques. We can generate highly detailed avatars with realistic hairstyles and facial hair. We also demonstrate 3D avatar generation from image or text, as well as text-guided editability.

We present a high-fidelity 3D generative adversarial network (GAN) inversion framework that can synthesize photorealistic novel views while preserving specific details of the input image. High-fidelity 3D GAN inversion is inherently challenging due to the geometry-texture trade-off, where overfitting to a single view input image often damages the estimated geometry during the latent optimization. To solve this challenge, we propose a novel pipeline that builds on the pseudo-multi-view estimation with visibility analysis. We keep the original textures for the visible parts and utilize generative priors for the occluded parts. Extensive experiments show that our approach achieves advantageous reconstruction and novel view synthesis quality over prior work, even for images with out-of-distribution textures. The proposed pipeline also enables image attribute editing with the inverted latent code and 3D-aware texture modification. Our approach enables high-fidelity 3D rendering from a single image, which is promising for various applications of AI-generated 3D content. The source code is at https://github.com/jiaxinxie97/HFGI3D/.

In this work, we propose an ID-preserving talking head generation framework, which advances previous methods in two aspects. First, as opposed to interpolating from sparse flow, we claim that dense landmarks are crucial to achieving accurate geometry-aware flow fields. Second, inspired by face-swapping methods, we adaptively fuse the source identity during synthesis, so that the network better preserves the key characteristics of the image portrait. Although the proposed model surpasses prior generation fidelity on established benchmarks, personalized fine-tuning is still needed to further make the talking head generation qualified for real usage. However, this process is rather computationally demanding that is unaffordable to standard users. To alleviate this, we propose a fast adaptation model using a metalearning approach. The learned model can be adapted to a high-quality personalized model as fast as 30 seconds. Last but not least, a spatial-temporal enhancement module is proposed to improve the fine details while ensuring temporal coherency. Extensive experiments prove the significant superiority of our approach over the state of the arts in both one-shot and personalized settings.

Self-supervised representation learning follows a paradigm of withholding some part of the data and tasking the network to predict it from the remaining part. Among many techniques, data augmentation lies at the core for creating the information gap. Towards this end, masking has emerged as a generic and powerful tool where content is withheld along the sequential dimension, e.g., spatial in images, temporal in audio, and syntactic in language. In this paper, we explore the orthogonal channel dimension for generic data augmentation by exploiting precision redundancy. The data for each channel is quantized through a non-uniform quantizer, with the quantized value sampled randomly within randomly sampled quantization bins. From another perspective, quantization is analogous to channel-wise masking, as it removes the information within each bin, but preserves the information across bins. Our approach significantly surpasses existing generic data augmentation methods, while showing on par performance against modality-specific augmentations. We comprehensively evaluate our approach on vision, audio, 3D point clouds, as well as the DABS benchmark which is comprised of various data modalities. The code is available at https://github.com/microsoft/random_quantize.

The diffusion-based generative models have achieved remarkable success in text-based image generation. However, since it contains enormous randomness in generation progress, it is still challenging to apply such models for real-world visual content editing, especially in videos. In this paper, we propose FateZero, a zero-shot text-based editing method on real-world videos without per-prompt training or use-specific mask. To edit videos consistently, we propose several techniques based on the pre-trained models. Firstly, in contrast to the straightforward DDIM inversion technique, our approach captures intermediate attention maps during inversion, which effectively retain both structural and motion information. These maps are directly fused in the editing process rather than generated during denoising. To further minimize semantic leakage of the source video, we then fuse self-attentions with a blending mask obtained by cross-attention features from the source prompt. Furthermore, we have implemented a reform of the self-attention mechanism in denoising UNet by introducing spatial-temporal attention to ensure frame consistency. Yet succinct, our method is the first one to show the ability of zero-shot text-driven video style and local attribute editing from the trained text-to-image model. We also have a better zero-shot shape-aware editing ability based on the text-to-video model [52]. Extensive experiments demonstrate our superior temporal consistency and editing capability than previous works.

We present a novel framework to bootstrap Motion forecastIng with Self-consistent Constraints (MISC). The motion forecasting task aims at predicting future trajectories of vehicles by incorporating spatial and temporal information from the past. A key design of MISC is the proposed Dual Consistency Constraints that regularize the predicted trajectories under spatial and temporal perturbation during training. Also, to model the multi-modality in motion forecasting, we design a novel self-ensembling scheme to obtain accurate teacher targets to enforce the self-constraints with multi-modality supervision. With explicit constraints from multiple teacher targets, we observe a clear improvement in the prediction performance. Extensive experiments on the Argoverse motion forecasting benchmark and Waymo Open Motion dataset show that MISC significantly outperforms the state-of-the-art methods. As the proposed strategies are general and can be easily incorporated into other motion forecasting approaches, we also demonstrate that our proposed scheme consistently improves the prediction performance of several existing methods.

This work presents an easy-to-use regularizer for GAN training, which helps explicitly link some axes of the latent space to a set of pixels in the synthesized image. Establishing such a connection facilitates a more convenient local control of GAN generation, where users can alter the image content only within a spatial area simply by partially resampling the latent code. Experimental results confirm four appealing properties of our regularizer, which we call LinkGAN. (1) The latent-pixel linkage is applicable to either a fixed region (i.e., same for all instances) or a particular semantic category (i.e., varying across instances), like the sky. (2) Two or multiple regions can be independently linked to different latent axes, which further supports joint control. (3) Our regularizer can improve the spatial controllability of both 2D and 3D-aware GAN models, barely sacrificing the synthesis performance. (4) The models trained with our regularizer are compatible with GAN inversion techniques and maintain editability on real images. Project page can be found here.

In robotic applications, we often obtain tons of 3D point cloud data without color information, and it is difficult to visualize point clouds in a meaningful and colorful way. Can we colorize 3D point clouds for better visualization? Existing deep learning-based colorization methods usually only take simple 3D objects as input, and their performance for complex scenes with multiple objects is limited. To this end, this paper proposes a novel semantics-and-geometry-aware colorization network, termed SGNet, for vivid scene-level point cloud colorization. Specifically, we propose a novel pipeline that explores geometric and semantic cues from point clouds containing only coordinates for color prediction. We also design two novel losses, including a colorfulness metric loss and a pairwise consistency loss, to constrain model training for genuine colorization. To the best of our knowledge, our work is the first to generate realistic colors for point clouds of large-scale indoor scenes. Extensive experiments on the widely used ScanNet benchmarks demonstrate that the proposed method achieves state-of-the-art performance on point cloud colorization.

Existing video super-resolution methods often utilize a few neighboring frames to generate a higher-resolution image for each frame. However, the abundant information in distant frames has not been fully exploited in these methods: corresponding patches of the same instance appear across distant frames at different scales. Based on this observation, we propose to improve the video super-resolution quality with long-term cross-scale aggregation that leverages similar patches (self-exemplars) across distant frames. Our method can be implemented as post-processing for any super-resolution methods to improve performance. Our model consists of a multi-reference alignment module to fuse the features derived from similar patches: we fuse the features of distant references to perform high-quality super-resolution. We also propose a novel and practical training strategy for reference-based super-resolution. To evaluate the performance of our proposed method, we conduct extensive experiments on our collected CarCam dataset, the Waymo Open dataset, and the REDS dataset, and the results demonstrate our method outperforms state-of-the-art methods.

The waterdrops on windshields during driving can cause severe visual obstructions, which may lead to car accidents. Meanwhile, the waterdrops can also degrade the performance of a computer vision system in autonomous driving. To address these issues, we propose an attention-based framework that fuses the spatio-temporal representations from multiple frames to restore visual information occluded by waterdrops. Due to the lack of training data for video waterdrop removal, we propose a large-scale synthetic dataset with simulated waterdrops in complex driving scenes on rainy days. To improve the generality of our proposed method, we adopt a cross-modality training strategy that combines synthetic videos and real-world images. Extensive experiments show that our proposed method can generalize well and achieve the best waterdrop removal performance in complex real-world driving scenes.

This paper tackles the task of singulating and grasping paper-like deformable objects. We refer to such tasks as paper-flipping. In contrast to manipulating deformable objects that lack compression strength (such as shirts and ropes), minor variations in the physical properties of the paper-like deformable objects significantly impact the results, making manipulation highly challenging. Here, we present Flipbot, a novel solution for flipping paper-like deformable objects. Flipbot allows the robot to capture object physical properties by integrating exteroceptive and proprioceptive perceptions that are indispensable for manipulating deformable objects. Furthermore, by incorporating a proposed coarse-to-fine exploration process, the system is capable of learning the optimal control parameters for effective paper-flipping through proprioceptive and exteroceptive inputs. We deploy our method on a real-world robot with a soft gripper and learn in a self-supervised manner. The resulting policy demonstrates the effectiveness of Flipbot on paper-flipping tasks with various settings beyond the reach of prior studies, including but not limited to flipping pages throughout a book and emptying paper sheets in a box. The code is available here: https://robotll.github.io/Flipbot/.

Since the advent of social media platforms, image selection based on social preference is a challenging task that all users inherently undertake before sharing images with the public. In our user study for this problem, human choices of images based on perceived social preference are largely inaccurate (58.7% accuracy). The challenge of this task, also known as image popularity assessment, lies in its subjective nature caused by visual and non-visual factors. Especially in the social media setting, social feedback on a particular image largely differs depending on who uploads it. Therefore social preference model should be able to account for this user-specific image aspect of the task. To address this issue, we present a retrieval-augmented approach that leverages both image features and user-specific statistics for neural image popularity assessment. User-specific statistics are derived by retrieving past images with their statistics from a memory bank. By combining these statistics with image features, our approach achieves 79.5% accuracy, which significantly outperforms human and baseline models on the pairwise ranking of images from the Instagram Influencer Dataset. Our source code will be publicly available.

Diffusion models have gained increasing attention for their impressive generation abilities but currently struggle with rendering accurate and coherent text. To address this issue, we introduce TextDiffuser, focusing on generating images with visually appealing text that is coherent with backgrounds. TextDiffuser consists of two stages: first, a Transformer model generates the layout of keywords extracted from text prompts, and then diffusion models generate images conditioned on the text prompt and the generated layout. Additionally, we contribute the first large-scale text images dataset with OCR annotations, MARIO-10M, containing 10 million image-text pairs with text recognition, detection, and character-level segmentation annotations. We further collect the MARIO-Eval benchmark to serve as a comprehensive tool for evaluating text rendering quality. Through experiments and user studies, we demonstrate that TextDiffuser is flexible and controllable to create high-quality text images using text prompts alone or together with text template images, and conduct text inpainting to reconstruct incomplete images with text. We will make the code, model and dataset publicly available.

We are living in a three-dimensional space while moving forward through a fourth dimension: time. To allow artificial intelligence to develop a comprehensive understanding of such a 4D environment, we introduce **4D Panoptic Scene Graph (PSG-4D)**, a new representation that bridges the raw visual data perceived in a dynamic 4D world and high-level visual understanding. Specifically, PSG-4D abstracts rich 4D sensory data into nodes, which represent entities with precise location and status information, and edges, which capture the temporal relations. To facilitate research in this new area, we build a richly annotated PSG-4D dataset consisting of 3K RGB-D videos with a total of 1M frames, each of which is labeled with 4D panoptic segmentation masks as well as fine-grained, dynamic scene graphs. To solve PSG-4D, we propose PSG4DFormer, a Transformer-based model that can predict panoptic segmentation masks, track masks along the time axis, and generate the corresponding scene graphs via a relation component. Extensive experiments on the new dataset show that our method can serve as a strong baseline for future research on PSG-4D. In the end, we provide a real-world application example to demonstrate how we can achieve dynamic scene understanding by integrating a large language model into our PSG-4D system.

Previous animatable 3D-aware GANs for human generation have primarily focused on either the human head or full body. However, head-only videos are relatively uncommon in real life, and full body generation typically does not deal with facial expression control and still has challenges in generating high-quality results. Towards applicable video avatars, we present an animatable 3D-aware GAN that generates portrait images with controllable facial expression, head pose, and shoulder movements. It is a generative model trained on unstructured 2D image collections without using 3D or video data. For the new task, we base our method on the generative radiance manifold representation and equip it with learnable facial and head-shoulder deformations. A dual-camera rendering and adversarial learning scheme is proposed to improve the quality of the generated faces, which is critical for portrait images. A pose deformation processing network is developed to generate plausible deformations for challenging regions such as long hair. Experiments show that our method, trained on unstructured 2D images, can generate diverse and high-quality 3D portraits with desired control over different properties.

Domain generalization (DG) has been a hot topic in image recognition, with a goal to train a general model that can perform well on unseen domains. Recently, federated learning (FL), an emerging machine learning paradigm to train a global model from multiple decentralized clients without compromising data privacy, has brought new challenges and possibilities to DG. In the FL scenario, many existing state-of-the-art (SOTA) DG methods become ineffective because they require the centralization of data from different domains during training. In this paper, we propose a novel domain generalization method for image recognition under federated learning through cross-client style transfer (CCST) without exchanging data samples. Our CCST method can lead to more uniform distributions of source clients, and make each local model learn to fit the image styles of all the clients to avoid the different model biases. Two types of style (single image style and overall domain style) with corresponding mechanisms are proposed to be chosen according to different scenarios. Our style representation is exceptionally lightweight and can hardly be used to reconstruct the dataset. The level of diversity is also flexible to be controlled with a hyper-parameter. Our method outperforms recent SOTA DG methods on two DG benchmarks (PACS, OfficeHome) and a large-scale medical image dataset (Camelyon17) in the FL setting. Last but not least, our method is orthogonal to many classic DG methods, achieving additive performance by combined utilization. Our code is available at: https://chenjunming.ml/proj/CCST.

We propose a simple yet effective reflection-free cue for robust reflection removal from a pair of flash and ambient (no-flash) images. The reflection-free cue exploits a flash-only image obtained by subtracting the ambient image from the corresponding flash image in raw data space. The flash-only image is equivalent to an image taken in a dark environment with only a flash on. This flash-only image is visually reflection-free and thus can provide robust cues to infer the reflection in the ambient image. Since the flash-only image usually has artifacts, we further propose a dedicated model that not only utilizes the reflection-free cue but also avoids introducing artifacts, which helps accurately estimate reflection and transmission. Our experiments on real-world images with various types of reflection demonstrate the effectiveness of our model with reflection-free flash-only cues: our model outperforms state-of-the-art reflection removal approaches by more than 5.23 dB in PSNR. We extend our approach to handheld photography to address the misalignment between the flash and no-flash pair. With misaligned training data and the alignment module, our aligned model outperforms our previous version by more than 3.19 dB in PSNR on a misaligned dataset. We also study using linear RGB images as training data.

Applying an image processing algorithm independently to each video frame often leads to temporal inconsistency in the resulting video. To address this issue, we present a novel and general approach for blind video temporal consistency. Our method is only trained on a pair of original and processed videos directly instead of a large dataset. Unlike most previous methods that enforce temporal consistency with optical flow, we show that temporal consistency can be achieved by training a convolutional network on a video with Deep Video Prior (DVP). Moreover, a carefully designed iteratively reweighted training strategy is proposed to address the challenging multimodal inconsistency problem. We demonstrate the effectiveness of our approach on 7 computer vision tasks on videos. Extensive quantitative and perceptual experiments show that our approach obtains superior performance than state-of-the-art methods on blind video temporal consistency. We further extend DVP to video propagation and demonstrate its effectiveness in propagating three different types of information (color, artistic style, and object segmentation). A progressive propagation strategy with pseudo labels is also proposed to enhance DVP's performance on video propagation. Our source codes are publicly available at https://github.com/ChenyangLEI/deep-video-prior.

Image operators have been extensively applied to create visually attractive photos for users to share processed images on social media. However, most image operators often smooth out details or generate textures after the processing, which removes the original content and raises challenges for restoring the original image. To resolve this issue, we propose a quasi-invertible model that learns common image processing operators in a restorable fashion: the learned image operators can generate visually pleasing results with the original content embedded. Our model is trained on input-output pairs that represent an image processing operator's behavior and uses a network that consists of an invertible branch and a non-invertible branch to increase our model's approximation capability. We evaluate the proposed model on ten image operators, including detail enhancement, abstraction, blur, photographic style, and non-photorealistic style. Extensive experiments show that our approach outperforms relevant baselines in the restoration quality, and the learned restorable operator is fast in inference and robust to compression. Furthermore, we demonstrate that the invertible operator can be easily applied to practical applications such as restorable human face retouching and highlight preserved exposure adjustment.

Video prediction is an extrapolation task that predicts future frames given past frames, and video frame interpolation is an interpolation task that estimates intermediate frames between two frames. We have witnessed the tremendous advancement of video frame interpolation, but the general video prediction in the wild is still an open question. Inspired by the photo-realistic results of video frame interpolation, we present a new optimization framework for video prediction via video frame interpolation, in which we solve an extrapolation problem based on an interpolation model. Our video prediction framework is based on optimization with a pretrained differentiable video frame interpolation module without the need for a training dataset, and thus there is no domain gap issue between training and test data. Also, our approach does not need any additional information such as semantic or instance maps, which makes our framework applicable to any video. Extensive experiments on the Cityscapes, KITTI, DAVIS, Middlebury, and Vimeo90K datasets show that our video prediction results are robust in general scenarios, and our approach outperforms other video prediction methods that require a large amount of training data or extra semantic information.

6D object pose estimation networks are limited in their capability to scale to large numbers of object instances due to the close-set assumption and their reliance on high-fidelity object CAD models. In this work, we study a new open set problem; the few-shot 6D object poses estimation: estimating the 6D pose of an unknown object by a few support views without extra training. To tackle the problem, we point out the importance of fully exploring the appearance and geometric relationship between the given support views and query scene patches and propose a dense prototypes matching framework by extracting and matching dense RGBD prototypes with transformers. Moreover, we show that the priors from diverse appearances and shapes are crucial to the generalization capability under the problem setting and thus propose a large-scale RGBD photorealistic dataset (ShapeNet6D) for network pre-training. A simple and effective online texture blending approach is also introduced to eliminate the domain gap from the synthesis dataset, which enriches appearance diversity at a low cost. Finally, we discuss possible solutions to this problem and establish benchmarks on popular datasets to facilitate future research. [project page]

We present a new data-driven approach with physics based priors to scene-level normal estimation from a single polarization image. Existing shape from polarization (SfP) works mainly focus on estimating the normal of a single object rather than complex scenes in the wild. A key barrier to high-quality scene-level SfP is the lack of real-world SfP data in complex scenes. Hence, we contribute the first real world scene-level SfP dataset with paired input polarization images and ground-truth normal maps. Then we propose a learning-based framework with a multi-head self-attention module and viewing encoding, which is designed to handle increasing polarization ambiguities caused by complex materials and non-orthographic projection in scene-level SfP. Our trained model can be generalized to far-field outdoor scenes as the relationship between polarized light and surface normals is not affected by distance. Experimental results demonstrate that our approach significantly outperforms existing SfP models on two datasets. Our dataset and source code will be publicly available at https://github.com/ChenyangLEI/sfp-wild.

We present a novel highfidelity generative adversarial network (GAN) inversion framework that enables attribute editing with image-specific details well-preserved (e.g., background, appearance, and illumination). We first analyze the challenges of highfidelity GAN inversion from the perspective of lossy data compression. With a low bitrate latent code, previous works have difficulties in preserving highfidelity details in reconstructed and edited images. Increasing the size of a latent code can improve the accuracy of GAN inversion but at the cost of inferior editability. To improve image fidelity without compromising editability, we propose a distortion consultation approach that employs a distortion map as a reference for highfidelity reconstruction. In the distortion consultation inversion (DCI), the distortion map is first projected to a high-rate latent map, which then complements the basic low-rate latent code with more details via consultation fusion. To achieve high-fidelity editing, we propose an adaptive distortion alignment (ADA) module with a self-supervised training scheme, which bridges the gap between the edited and inversion images. Extensive experiments in the face and car domains show a clear improvement in both inversion and editing quality. The project page is https://tengfei-wang.github.io/HFGI/.

The degenerate unmixing estimation technique (DUET) is one of the most efficient blind source separation algorithms tackling the challenging situation when the number of sources exceeds the number of microphones. However, as a time-frequency mask-based method, DUET erroneously results in interference components retention when source signals overlap each other in both frequency and time domains. In this paper, to avoid the erroneous retention, instead of masking, we propose to use multiple linear spatial filters (e.g., the minimum variance distortionless response filter) to extract the desired signals. These filters are constructed based on the information embedded in the detected single-source-points, that is, time-frequency points contributed by a single source. In comparison with the conventional DUET, our method achieved an impressive improvement greater than 5 dB in the source-to-interference ratio and 2 to 5 dB improvement in the source-to-distortion ratio, respectively. Findings are substantiated by unmixing simulation using live-recorded mixture signals from up to four sources. Audio examples can be found on the web page: "https://ydcnanhe.github.io/demoicassp2022/"

Despite the rapid advancement of semantic discovery in the latent space of Generative Adversarial Networks (GANs), existing approaches either are limited to finding global attributes or rely on a number of segmentation masks to identify local attributes. In this work, we present a highly efficient algorithm to factorize the latent semantics learned by GANs concerning an arbitrary image region. Concretely, we revisit the task of local manipulation with pre-trained GANs and formulate region-based semantic discovery as a dual optimization problem. Through an appropriately defined generalized Rayleigh quotient, we manage to solve such a problem without any annotations or training. Experimental results on various state-of-the-art GAN models demonstrate the effectiveness of our approach, as well as its superiority over prior arts regarding precise control, region robustness, speed of implementation, and simplicity of use.

We present a portable multiscopic camera system with a dedicated model for novel view and time synthesis in dynamic scenes. Our goal is to render high-quality images for a dynamic scene from any viewpoint at any time using our portable multiscopic camera. To achieve such novel view and time synthesis, we develop a physical multiscopic camera equipped with five cameras to train a neural radiance field (NeRF) in both time and spatial domains for dynamic scenes. Our model maps a 6D coordinate (3D spatial position, 1D temporal coordinate, and 2D viewing direction) to view-dependent and time-varying emitted radiance and volume density. Volume rendering is applied to render a photo-realistic image at a specified camera pose and time. To improve the robustness of our physical camera, we propose a camera parameter optimization module and a temporal frame interpolation module to promote information propagation across time. We conduct experiments on both real-world and synthetic datasets to evaluate our system, and the results show that our approach outperforms alternative solutions qualitatively and quantitatively. Our code and dataset are available at https://yuenfuilau.github.io/.

With the rise of deep learning and intelligent vehicles, the smart assistant has become an essential in-car component to facilitate driving and provide extra functionalities. In-car smart assistants should be able to process general as well as car-related commands and perform corresponding actions, which eases driving and improves safety. However, there is a data scarcity issue for low resource languages, hindering the development of research and applications. In this paper, we introduce a new dataset, Cantonese In-car Audio-Visual Speech Recognition (CI-AVSR), for in-car command recognition in the Cantonese language with both video and audio data. It consists of 4,984 samples (8.3 hours) of 200 in-car commands recorded by 30 native Cantonese speakers. Furthermore, we augment our dataset using common in-car background noises to simulate real environments, producing a dataset 10 times larger than the collected one. We provide detailed statistics of both the clean and the augmented versions of our dataset. Moreover, we implement two multimodal baselines to demonstrate the validity of CI-AVSR. Experiment results show that leveraging the visual signal improves the overall performance of the model. Although our best model can achieve a considerable quality on the clean test set, the speech recognition quality on the noisy data is still inferior and remains an extremely challenging task for real in-car speech recognition systems. The dataset and code will be released at https://github.com/HLTCHKUST/CI-AVSR.

Code-switching is a speech phenomenon occurring when a speaker switches language during a conversation. Despite the spontaneous nature of code-switching in conversational spoken language, most existing works collect code-switching data from read speech instead of spontaneous speech. ASCEND (A Spontaneous Chinese-English Dataset) is a high-quality Mandarin Chinese-English code-switching corpus built on spontaneous multi-turn conversational dialogue sources collected in Hong Kong. We report ASCEND’s design and procedure for collecting the speech data, including annotations. ASCEND consists of 10.62 hours of clean speech, collected from 23 bilingual speakers of Chinese and English. Furthermore, we conduct baseline experiments using pre-trained wav2vec 2.0 models, achieving a best performance of 22.69% character error rate and 27.05% mixed error rate.

Automatic speech recognition (ASR) on low resource languages improves the access of linguistic minorities to technological advantages provided by artificial intelligence (AI). In this paper, we address the problem of data scarcity for the Hong Kong Cantonese language by creating a new Cantonese dataset. Our dataset, Multi-Domain Cantonese Corpus (MDCC), consists of 73.6 hours of clean read speech paired with transcripts, collected from Cantonese audiobooks from Hong Kong. It comprises philosophy, politics, education, culture, lifestyle and family domains, covering a wide range of topics. We also review all existing Cantonese datasets and analyze them according to their speech type, data source, total size and availability. We further conduct experiments with Fairseq S2T Transformer, a state-of-the-art ASR model, on the biggest existing dataset, Common Voice zh-HK, and our proposed MDCC, and the results show the effectiveness of our dataset. In addition, we create a powerful and robust Cantonese ASR model by applying multi-dataset learning on MDCC and Common Voice zh-HK.

Video streams are delivered continuously to save the cost of storage and device memory. Real-time denoising algorithms are typically adopted on the user device to remove the noise involved during the shooting and transmission of video streams. However, sliding-window-based methods feed multiple input frames for a single output and lack computation efficiency. Recent multi-output inference works propagate the bidirectional temporal feature with a parallel or recurrent framework, which either suffers from performance drops on the temporal edges of clips or can not achieve online inference. In this paper, we propose a Bidirectional Streaming Video Denoising (BSVD) framework, to achieve high-fidelity real-time denoising for streaming videos with both past and future temporal receptive fields. The bidirectional temporal fusion for online inference is considered not applicable in the MoViNet. However, we introduce a novel Bidirectional Buffer Block as the core module of our BSVD, which makes it possible during our pipeline-style inference. In addition, our method is concise and flexible to be utilized in both non-blind and blind video denoising. We compare our model with various state-of-the-art video denoising models qualitatively and quantitatively on synthetic and real noise. Our method outperforms previous methods in terms of restoration fidelity and runtime.

Compositing portrait photographs or videos to novel backgrounds is an important application in computational photography. Seamless blending along boundaries and globally harmonic colors are two desired properties of the photo-realistic composition of foregrounds and new backgrounds. Existing works are dedicated to either foreground alpha matte generation or after-blending harmonization, leading to sub-optimal background replacement when putting foregrounds and backgrounds together. In this work, we unify the two objectives in a single framework to obtain realistic portrait image composites. Specifically, we investigate the usage of a target background and find that a complete background plays a vital role in both seamlessly blending and harmonization. We develop a network to learn the composition process given an imperfect alpha matte with appearance features extracted from the complete background to adjust color distribution. Our dedicated usage of a complete background enables realistic portrait image composition and also temporally stable results on videos. Extensive quantitative and qualitative experiments on both synthetic and real-world data demonstrate that our method achieves state-of-the-art performance.

Although 2D generative models have made great progress in face image generation and animation, they often suffer from undesirable artifacts such as 3D inconsistency when rendering images from different camera viewpoints. This prevents them from synthesizing video animations indistinguishable from real ones. Recently, 3D-aware GANs extend 2D GANs for explicit disentanglement of camera pose by leveraging 3D scene representations. These methods can well preserve the 3D consistency of the generated images across different views, yet they cannot achieve fine-grained control over other attributes, among which facial expression control is arguably the most useful and desirable for face animation. In this paper, we propose an animatable 3D-aware GAN for multiview consistent face animation generation. The key idea is to decompose the 3D representation of the 3D-aware GAN into a template field and a deformation field, where the former represents different identities with a canonical expression, and the latter characterizes expression variations of each identity. To achieve meaningful control over facial expressions via deformation, we propose a 3D-level imitative learning scheme between the generator and a parametric 3D face model during adversarial training of the 3D-aware GAN. This helps our method achieve high-quality animatable face image generation with strong visual 3D consistency, even though trained with only unstructured 2D images. Extensive experiments demonstrate our superior performance over prior works. Project page: \url{https://yuewuhkust.github.io/AniFaceGAN/

3D-aware image synthesis aims at learning a generative model that can render photo-realistic 2D images while capturing decent underlying 3D shapes. A popular solution is to adopt the generative adversarial network (GAN) and replace the generator with a 3D renderer, where volume rendering with neural radiance field (NeRF) is commonly used. Despite the advancement of synthesis quality, existing methods fail to obtain moderate 3D shapes. We argue that, considering the two-player game in the formulation of GANs, only making the generator 3D-aware is not enough. In other words, displacing the generative mechanism only offers the capability, but not the guarantee, of producing 3D-aware images, because the supervision of the generator primarily comes from the discriminator. To address this issue, we propose GeoD through learning a geometry-aware discriminator to improve 3D-aware GANs. Concretely, besides differentiating real and fake samples from the 2D image space, the discriminator is additionally asked to derive the geometry information from the inputs, which is then applied as the guidance of the generator. Such a simple yet effective design facilitates learning substantially more accurate 3D shapes. Extensive experiments on various generator architectures and training datasets verify the superiority of GeoD over state-of-the-art alternatives. Moreover, our approach is registered as a general framework such that a more capable discriminator (i.e., with a third task of novel view synthesis beyond domain classification and geometry extraction) can further assist the generator with a better multi-view consistency. Project page can be found at https://vivianszf.github.io/geod.

Imitation learning is a class of promising policy learning algorithms that is free from many practical issues with reinforcement learning, such as the reward design issue and the exploration hardness. However, the current imitation algorithm struggles to achieve both high performance and high in-environment sample efficiency simultaneously. Behavioral Cloning (BC) does not need in-environment interactions, but it suffers from the covariate shift problem which harms its performance. Adversarial Imitation Learning (AIL) turns imitation learning into a distribution matching problem. It can achieve better performance on some tasks but it requires a large number of in-environment interactions. Inspired by the recent success of EfficientZero in RL, we propose EfficientImitate (EI), a planning-based imitation learning method that can achieve high in-environment sample efficiency and performance simultaneously. Our algorithmic contribution in this paper is two-fold. First, we extend AIL into the MCTS-based RL. Second, we show the seemingly incompatible two classes of imitation algorithms (BC and AIL) can be naturally unified under our framework, enjoying the benefits of both. We benchmark our method not only on the state-based DeepMind Control Suite but also on the image version which many previous works find highly challenging. Experimental results show that EI achieves state-of-the-art results in performance and sample efficiency. EI shows over 4x gain in performance in the limited sample setting on state-based and image-based tasks and can solve challenging problems like Humanoid, where previous methods fail with a small amount of interactions. Our code is available at https://github.com/zhaohengyin/EfficientImitate.

Free-form text prompts allow users to describe their intentions during image manipulation conveniently. Based on the visual latent space of StyleGAN[21] and text embedding space of CLIP[34], studies focus on how to map these two latent spaces for text-driven attribute manipulations. Currently, the latent mapping between these two spaces is empirically designed and confines that each manipulation model can only handle one fixed text prompt. In this paper, we propose a method named Free-Form CLIP (FFCLIP), aiming to establish an automatic latent mapping so that one manipulation model handles free-form text prompts. Our FFCLIP has a cross-modality semantic modulation module containing semantic alignment and injection. The semantic alignment performs the automatic latent mapping via linear transformations with a cross attention mechanism. After alignment, we inject semantics from text prompt embeddings to the StyleGAN latent space. For one type of image (e.g., human portrait'), one FFCLIP model can be learned to handle free-form text prompts. Meanwhile, we observe that although each training text prompt only contains a single semantic meaning, FFCLIP can leverage text prompts with multiple semantic meanings for image manipulation. In the experiments, we evaluate FFCLIP on three types of images (i.e.,human portraits', cars', andchurches'). Both visual and numerical results show that FFCLIP effectively produces semantically accurate and visually realistic images. Project page: https://github.com/KumapowerLIU/FFCLIP.

Although recent inpainting approaches have demonstrated significant improvement with deep neural networks, they still suffer from artifacts such as blunt structures and abrupt colors when filling in the missing regions. To address these issues, we propose an external-internal inpainting scheme with a monochromic bottleneck that helps image inpainting models remove these artifacts. In the external learning stage, we reconstruct missing structures and details in the monochromic space to reduce the learning dimension. In the internal learning stage, we propose a novel internal color propagation method with progressive learning strategies for consistent color restoration. Extensive experiments demonstrate that our proposed scheme helps image inpainting models produce more structure-preserved and visually compelling results.

In this work, we present FFB6D, a Full Flow Bidirectional fusion network designed for 6D pose estimation from a single RGBD image. Our key insight is that appearance information in the RGB image and geometry information from the depth image are two complementary data sources, and it still remains unknown how to fully leverage them. Towards this end, we propose FFB6D, which learns to combine appearance and geometry information for representation learning as well as output representation selection. Specifically, at the representation learning stage, we build bidirectional fusion modules in the full flow of the two networks, where fusion is applied to each encoding and decoding layer. In this way, the two networks can leverage local and global complementary in-formation from the other one to obtain better representations. Moreover, at the output representation stage, we designed a simple but effective 3D keypoints selection algorithm considering the texture and geometry information of objects, which simplifies keypoint localization for precise pose estimation. Experimental results show that our method outperforms the state-of-the-art by large margins on several benchmarks. Code and video are available at https://github.com/ethnhe/FFB6D.git.

We propose a simple yet effective reflection-free cue for robust reflection removal from a pair of flash and ambient (no-flash) images. The reflection-free cue exploits a flash-only image obtained by subtracting the ambient image from the corresponding flash image in raw data space. The flash-only image is equivalent to an image taken in a dark environment with only a flash on. We observe that this flash-only image is visually reflection-free, and thus it can provide robust cues to infer the reflection in the ambient image. Since the flash-only image usually has artifacts, we further propose a dedicated model that not only utilizes the reflection-free cue but also avoids introducing artifacts, which helps accurately estimate reflection and transmission. Our experiments on real-world images with various types of reflection demonstrate the effectiveness of our model with reflection-free flash-only cues: our model outperforms state-of-the-art reflection removal approaches by more than 5.23dB in PSNR, 0.04 in SSIM, and 0.068 in LPIPS. Our source code and dataset are publicly available at github.com/ChenyangLEI/flash-reflection-removal.

Convolution has been the core ingredient of modern neural networks, triggering the surge of deep learning in vision. In this work, we rethink the inherent principles of standard convolution for vision tasks, specifically spatial-agnostic and channel-specific. Instead, we present a novel atomic operation for deep neural networks by inverting the aforementioned design principles of convolution, coined as involution. We additionally demystify the recent popular self-attention operator and subsume it into our involution family as an over-complicated instantiation. The proposed involution operator could be leveraged as fundamental bricks to build the new generation of neural networks for visual recognition, powering different deep learning models on several prevalent benchmarks, including ImageNet classification, COCO detection and segmentation, together with Cityscapes segmentation. Our involution-based models improve the performance of convolutional baselines using ResNet-50 by up to 1.6% top-1 accuracy, 2.5% and 2.4% bounding box AP, and 4.7% mean IoU absolutely while compressing the computational cost to 66%, 65%, 72%, and 57% on the above benchmarks, respectively. Code and pre-trained models for all the tasks are available at https://github.com/d-li14/involution.

We present a controllable camera simulator based on deep neural networks to synthesize raw image data under different camera settings, including exposure time, ISO, and aperture. The proposed simulator includes an exposure module that utilizes the principle of modern lens designs for correcting the luminance level. It also contains a noise module using the noise level function and an aperture module with adaptive attention to simulate the side effects on noise and defocus blur. To facilitate the learning of a simulator model, we collect a dataset of the 10,000 raw images of 450 scenes with different exposure settings. Quantitative experiments and qualitative comparisons show that our approach outperforms relevant baselines in raw data synthesize on multiple cameras. Furthermore, the camera simulator enables various applications, including large-aperture enhancement, HDR, auto exposure, and data augmentation for training local feature detectors. Our work represents the first attempt to simulate a camera sensor’s behavior leveraging both the advantage of traditional raw sensor features and the power of data-driven deep learning. The code and the dataset are available at https://github.com/ken-ouyang/neural_image_simulator.

Unprocessed RAW data is a highly valuable image format for image editing and computer vision. However, since the file size of RAW data is huge, most users can only get access to processed and compressed sRGB images. To bridge this gap, we design an Invertible Image Signal Processing (InvISP) pipeline, which not only enables rendering visually appealing sRGB images but also allows recovering nearly perfect RAW data. Due to our framework’s inherent reversibility, we can reconstruct realistic RAW data instead of synthesizing RAW data from sRGB images without any memory overhead. We also integrate a differentiable JPEG compression simulator that empowers our framework to re-construct RAW data from JPEG images. Extensive quantitative and qualitative experiments on two DSLR demonstrate that our method obtains much higher quality in both rendered sRGB images and reconstructed RAW data than alternative methods.

We propose the Temporal Point Cloud Networks (TPCN), a novel and flexible framework with joint spatial and temporal learning for trajectory prediction. Unlike existing approaches that rasterize agents and map information as 2D images or operate in a graph representation, our approach extends ideas from point cloud learning with dynamic temporal learning to capture both spatial and temporal information by splitting trajectory prediction into both spatial and temporal dimensions. In the spatial dimension, agents can be viewed as an unordered point set, and thus it is straightforward to apply point cloud learning techniques to model agents’ locations. While the spatial dimension does not take kinematic and motion information into account, we further propose dynamic temporal learning to model agents’ motion over time. Experiments on the Argoverse motion forecasting benchmark show that our approach achieves state-of-the-art results.

We present a novel approach to reference-based super-resolution (RefSR) with the focus on dual-camera super-resolution (DCSR), which utilizes reference images for high-quality and high-fidelity results. Our proposed method generalizes the standard patch-based feature matching with spatial alignment operations. We further explore the dual-camera super-resolution that is one promising application of RefSR, and build a dataset that consists of 146 image pairs from the main and telephoto cameras in a smartphone. To bridge the domain gaps between real-world images and the training images, we propose a self-supervised domain adaptation strategy for real-world images. Extensive experiments on our dataset and a public benchmark demonstrate clear improvement achieved by our method over state of the art in both quantitative evaluation and visual comparisons. Our code and data are available at https://tengfei-wang.github.io/Dual-Camera-SR/index.html.

We propose a novel approach for embedding novel views in a single JPEG image while preserving the perceptual fidelity of the modified JPEG image and the restored novel views. We adopt the popular novel view synthesis representation of multiplane images (MPIs). Our model first encodes 32 MPI layers (totally 128 channels) into a 3-channel JPEG image that can be decoded for MPIs to render novel views, with an embedding capacity of 1024 bits per pixel. We conducted experiments on public datasets with different novel view synthesis methods, and the results show that the proposed method can restore high-fidelity novel views from a slightly modified JPEG image. Furthermore, our method is robust to JPEG compression, color adjusting, and cropping. Our source code will be publicly available.

Reversible image conversion (RIC) aims to build a reversible transformation between specific visual content (e.g., short videos) and an embedding image, where the original content can be restored from the embedding when necessary. This work develops Invertible Image Conversion Net (IIC-Net) as a generic solution to various RIC tasks due to its strong capacity and task-independent design. Unlike previous encoder-decoder based methods, IICNet maintains a highly invertible structure based on invertible neural networks (INNs) to better preserve the information during conversion. We use a relation module and a channel squeeze layer to improve the INN nonlinearity to extract cross-image relations and the network flexibility, respectively. Experimental results demonstrate that IICNet outperforms the specifically-designed methods on existing RIC tasks and can generalize well to various newly-explored tasks. With our generic IICNet, we no longer need to hand-engineer task-specific embedding networks for rapidly occurring visual content. Our source codes are available at: https://github.com/felixcheng97/IICNet.

We present a novel approach for extracting human pose features from human action videos. The goal is to let the pose features capture only the poses of the action while being invariant to other factors, including video back-grounds, the video subjects’ anthropometric characteristics and viewpoints. Such human pose features facilitate the comparison of pose similarity and can be used for down-stream tasks, such as human action video alignment and pose retrieval. The key to our approach is to first normalize the poses in the video frames by mapping the poses onto a pre-defined 3D skeleton to not only disentangle subject physical features, such as bone lengths and ratios, but also to unify global orientations of the poses. Then the normalized poses are mapped to a pose embedding space of high-level features, learned via unsupervised metric learning. We evaluate the effectiveness of our normalized features both qualitatively by visualizations, and quantitatively by a video alignment task on the Human3.6M dataset and an action recognition task on the Penn Action dataset.

We propose a novel framework for video inpainting by adopting an internal learning strategy. Unlike previous methods that use optical flow for cross-frame context propagation to inpaint unknown regions, we show that this can be achieved implicitly by fitting a convolutional neural network to known regions. Moreover, to handle challenging sequences with ambiguous backgrounds or long-term occlusion, we design two regularization terms to preserve high-frequency details and long-term temporal consistency. Extensive experiments on the DAVIS dataset demonstrate that the proposed method achieves state-of-the-art inpainting quality quantitatively and qualitatively. We further extend the proposed method to another challenging task: learning to remove an object from a video giving a single object mask in only one frame in a 4K video. Our source code is available at https://tengfei-wang.github.io/Implicit-Internal-Video-Inpainting/.

Motion prediction of vehicles is critical but challenging due to the uncertainties in complex environments and the limited visibility caused by occlusions and limited sensor ranges. In this paper, we study a new task, safety-aware motion prediction with unseen vehicles for autonomous driving. Unlike the existing trajectory prediction task for seen vehicles, we aim at predicting an occupancy map that indicates the earliest time when each location can be occupied by either seen and unseen vehicles. The ability to predict unseen vehicles is critical for safety in autonomous driving. To tackle this challenging task, we propose a safety-aware deep learning model with three new loss functions to predict the earliest occupancy map. Experiments on the large-scale autonomous driving nuScenes dataset show that our proposed model significantly outperforms the state-of-the-art baselines on the safety-aware motion prediction task. To the best of our knowledge, our approach is the first one that can predict the existence of unseen vehicles in most cases. Project page at https://github.com/xrenaa/Safety-Aware-Motion-Prediction.

We present a novel and flexible architecture for point cloud segmentation with dual-representation iterative learning. In point cloud processing, different representations have their own pros and cons. Thus, finding suitable ways to represent point cloud data structure while keeping its own internal physical property such as permutation and scale-invariant is a fundamental problem. Therefore, we propose our work, DRINet, which serves as the basic network structure for dual-representation learning with great flexibility at feature transferring and less computation cost, especially for large-scale point clouds. DRINet mainly consists of two modules called Sparse Point-Voxel Feature Extraction and Sparse Voxel-Point Feature Extraction. By utilizing these two modules iteratively, features can be propagated between two different representations. We further propose a novel multi-scale pooling layer for pointwise locality learning to improve context information propagation. Our network achieves state-of-the-art results for point cloud classification and segmentation tasks on several datasets while maintaining high runtime efficiency. For large-scale outdoor scenarios, our method outperforms state-of-the-art methods with a real-time inference time of 62ms per frame.

We propose SinIR, an efficient reconstruction-based framework trained on a single natural image for general image manipulation, including super-resolution, editing, harmonization, paint-to-image, photo-realistic style transfer, and artistic style transfer. We train our model on a single image with cascaded multi-scale learning, where each network at each scale is responsible for image reconstruction. This reconstruction objective greatly reduces the complexity and running time of training, compared to the GAN objective. However, the reconstruction objective also exacerbates the output quality. Therefore, to solve this problem, we further utilize simple random pixel shuffling, which also gives control over manipulation, inspired by the Denoising Autoencoder. With quantitative evaluation, we show that SinIR has competitive performance on various image manipulation tasks. Moreover, with a much simpler training objective (i.e., reconstruction), SinIR is trained 33.5 times faster than SinGAN (for 500x500 images) that solves similar tasks. Our code is publicly available at github.com/YooJiHyeong/SinIR.

We present a novel approach to joint depth and normal estimation for time-of-flight (ToF) sensors. Our model learns to predict the high-quality depth and normal maps jointly from ToF raw sensor data. To achieve this, we meticulously constructed the first large-scale dataset (named ToF-100) with paired raw ToF data and ground-truth high-resolution depth maps provided by an industrial depth camera. In addition, we also design a simple but effective framework for joint depth and normal estimation, applying a robust Chamfer loss via jittering to improve the performance of our model. Our experiments demonstrate that our proposed method can efficiently reconstruct high-resolution depth and normal maps and significantly outperforms state-of-the-art approaches.

Existing vision systems for autonomous driving or robots are sensitive to waterdrops adhered to windows or camera lenses. Most recent waterdrop removal approaches take a single image as input and often fail to recover the missing content behind waterdrops faithfully. Thus, we propose a learning-based model for waterdrop removal with stereo images. To better detect and remove waterdrops from stereo images, we propose a novel row-wise dilated attention module to enlarge attention’s receptive field for effective information propagation between the two stereo images. In addition, we propose an attention consistency loss between the ground-truth disparity map and attention scores to enhance the left-right consistency in stereo images. Because of related datasets’ unavailability, we collect a real-world dataset that contains stereo images with and without waterdrops. Extensive experiments on our dataset suggest that our model outperforms state-of-the-art methods both quantitatively and qualitatively. Our source code and the stereo waterdrop dataset are available at https://github.com/VivianSZF/Stereo-Waterdrop-Removal.

Self-supervised learning for depth estimation possesses several advantages over supervised learning. The benefits of no need for ground-truth depth, online fine-tuning, and better generalization with unlimited data attract researchers to seek self-supervised solutions. In this work, we propose a new self-supervised framework for stereo matching utilizing multiple images captured at aligned camera positions. A cross photometric loss, an uncertainty-aware mutual-supervision loss, and a new smoothness loss are introduced to optimize the network in learning disparity maps end-to-end without ground-truth depth information. To train this framework, we build a new multiscopic dataset consisting of synthetic images rendered by 3D engines and real images captured by real cameras. After being trained with only the synthetic images, our network can perform well in unseen outdoor scenes. Our experiment shows that our model obtains better disparity maps than previous unsupervised methods on the KITTI dataset and is comparable to supervised methods when generalized to unseen data. Our source code and dataset are available at https://sites.google.com/view/multiscopic.

Portrait images often suffer from undesirable shadows cast by casual objects or even the face itself. While existing methods for portrait shadow removal require training on a large-scale synthetic dataset, we propose the first unsupervised method for portrait shadow removal without any training data. Our key idea is to leverage the generative facial priors embedded in the off-the-shelf pretrained StyleGAN2. To achieve this, we formulate the shadow removal task as a layer decomposition problem: a shadowed portrait image is constructed by the blending of a shadow image and a shadow-free image. We propose an effective progressive optimization algorithm to learn the decomposition process. Our approach can also be extended to portrait tattoo removal and watermark removal. Qualitative and quantitative experiments on a real-world portrait shadow dataset demonstrate that our approach achieves comparable performance with supervised shadow removal methods. Our source code is available at https://github.com/YingqingHe/Shadow-Removal-via-Generative-Priors.

Although deep learning based image compression methods have achieved promising progress these days, the performance of these methods still cannot match the latest compression standard Versatile Video Coding (VVC). Most of the recent developments focus on designing a more accurate and flexible entropy model that can better parameterize the distributions of the latent features. However, few efforts are devoted to structuring a better transformation between the image space and the latent feature space. In this paper, instead of employing previous autoencoder style networks to build this transformation, we propose an enhanced Invertible Encoding Network with invertible neural networks (INNs) to largely mitigate the information loss problem for better compression. Experimental results on the Kodak, CLIC, and Tecnick datasets show that our method outperforms the existing learned image compression methods and compression standards, including VVC (VTM 12.1), especially for high-resolution images. Our source code is available at https://github.com/xyq7/InvCompress.

The latent space of a Generative Adversarial Network (GAN) has been shown to encode rich semantics within some subspaces. To identify these subspaces, researchers typically analyze the statistical information from a collection of synthesized data, and the identified subspaces tend to control image attributes globally (i.e., manipulating an attribute causes the change of an entire image). By contrast, this work introduces low-rank subspaces that enable more precise control of GAN generation. Concretely, given an arbitrary image and a region of interest (e.g., eyes of face images), we manage to relate the latent space to the image region with the Jacobian matrix and then use low-rank factorization to discover steerable latent subspaces. There are three distinguishable strengths of our approach that can be aptly called LowRankGAN. First, compared to analytic algorithms in prior work, our low-rank factorization of Jacobians is able to find the low-dimensional representation of attribute manifold, making image editing more precise and controllable. Second, low-rank factorization naturally yields a null space of attributes such that moving the latent code within it only affects the outer region of interest. Therefore, local image editing can be simply achieved by projecting an attribute vector into the null space without relying on a spatial mask as existing methods do. Third, our method can robustly work with a local region from one image for analysis yet well generalize to other images, making it much easy to use in practice. Extensive experiments on state-of-the-art GAN models (including StyleGAN2 and BigGAN) trained on various datasets demonstrate the effectiveness of our LowRankGAN.

We present a novel formulation to removing reflection from polarized images in the wild. We first identify the misalignment issues of existing reflection removal datasets where the collected reflection-free images are not perfectly aligned with input mixed images due to glass refraction. Then we build a new dataset with more than 100 types of glass in which obtained transmission images are perfectly aligned with input mixed images. Second, capitalizing on the special relationship between reflection and polarized light, we propose a polarized reflection removal model with a two-stage architecture. In addition, we design a novel perceptual NCC loss that can improve the performance of reflection removal and general image decomposition tasks. We conduct extensive experiments, and results suggest that our model outperforms state-of-the-art methods on reflection removal.

While the depth of modern Convolutional Neural Networks (CNNs) surpasses that of the pioneering networks with a significant margin, the traditional way of appending supervision only over the final classifier and progressively propagating gradient flow upstream remains the training mainstay. Seminal Deeply-Supervised Networks (DSN) were proposed to alleviate the difficulty of optimization arising from gradient flow through a long chain. However, it is still vulnerable to issues including interference to the hierarchical representation generation process and inconsistent optimization objectives, as illustrated theoretically and empirically in this paper. Complementary to previous training strategies, we propose Dynamic Hierarchical Mimicking, a generic feature learning mechanism, to advance CNN training with enhanced generalization ability. Partially inspired by DSN, we fork delicately designed side branches from the intermediate layers of a given neural network. Each branch can emerge from certain locations of the main branch dynamically, which not only retains representation rooted in the backbone network but also generates more diverse representations along its own pathway. We go one step further to promote multi-level interactions among different branches through an optimization formula with probabilistic prediction matching losses, thus guaranteeing a more robust optimization process and better representation ability. Experiments on both category and instance recognition tasks demonstrate the substantial improvements of our proposed method over its corresponding counterparts using diverse state-of-the-art CNN architectures. Code and models are publicly available at https://github.com/d-li14/DHM.

We present an approach to predict future video frames given a sequence of continuous video frames in the past. Instead of synthesizing images directly, our approach is designed to understand the complex scene dynamics by decoupling the background scene and moving objects. The appearance of the scene components in the future is predicted by non-rigid deformation of the background and affine transformation of moving objects. The anticipated appearances are combined to create a reasonable video in the future. With this procedure, our method exhibits much less tearing or distortion artifact compared to other approaches. Experimental results on the Cityscapes and KITTI datasets show that our model outperforms the state-of-the-art in terms of visual quality and accuracy.

Depth sensing is a critical component of autonomous driving technologies, but today's LiDAR- or stereo camera- based solutions have limited range. We seek to increase the maximum range of self-driving vehicles' depth perception modules for the sake of better safety. To that end, we propose a novel three-camera system that utilizes small field of view cameras. Our system, along with our novel algorithm for computing metric depth, does not require full pre-calibration and can output dense depth maps with practically acceptable accuracy for scenes and objects at long distances not well covered by most commercial LiDARs.

Depth from a monocular video can enable billions of devices and robots with a single camera to see the world in 3D. In this paper, we present a model for video depth estimation, which consists of a flow-to-depth layer, a camera pose refinement module, and a depth fusion network. Given optical flow and camera poses, our flow-to-depth layer generates depth proposals and their corresponding confidence maps by explicitly solving an epipolar geometry optimization problem. Our flow-to-depth layer is differentiable, and thus we can refine camera poses by maximizing the aggregated confidence in the camera pose refinement module. Our depth fusion network can utilize the target frame, depth proposals, and confidence maps inferred from different neighboring frames to produce the final depth map. Furthermore, the depth fusion network can additionally take the depth proposals generated by other methods to further improve the results. The experiments on three public datasets show that our approach outperforms state-of-the-art depth estimation methods, and has reasonable crossdataset generalization ability: our model trained on KITTI still performs well on the unseen Waymo dataset.

Self-supervised learning for visual object tracking possesses valuable advantages compared to supervised learning, such as the non-necessity of laborious human annotations and online training. In this work, we exploit an end-to-end Siamese network in a cycle-consistent self-supervised framework for object tracking. Self-supervision can be performed by taking advantage of the cycle consistency in the forward and backward tracking. To better leverage the end-to-end learning of deep networks, we propose to integrate a Siamese region proposal and mask regression network in our tracking framework so that a fast and more accurate tracker can be learned without the annotation of each frame. The experiments on the VOT dataset for visual object tracking and on the DAVIS dataset for video object segmentation propagation show that our method outperforms prior approaches on both tasks.

We present a self-supervised approach with pose perceptual loss for automatic dance video generation. Our method can produce a realistic dance video that conforms to the beats and rhymes of given music. To achieve this, we firstly generate a human skeleton sequence from music and then apply the learned pose-to-appearance mapping to generate the final video. In the stage of generating skeleton sequences, we utilize two discriminators to capture different aspects of the sequence and propose a novel pose perceptual loss to produce natural dances. Besides, we also provide a new cross-modal evaluation metric to evaluate the dance quality, which is able to estimate the similarity between two modalities (music and dance). Finally, our experimental qualitative and quantitative results demonstrate that our dance video synthesis approach produces realistic and diverse results. Our source code and data are available at https://github.com/xrenaa/Music-Dance-Video-Synthesis.

Applying image processing algorithms independently to each video frame often leads to temporal inconsistency in the resulting video. To address this issue, we present a novel and general approach for blind video temporal consistency. Our method is only trained on a pair of original and processed videos directly instead of a large dataset. Unlike most previous methods that enforce temporal consistency with optical flow, we show that temporal consistency can be achieved by training a convolutional network on a video with the Deep Video Prior. Moreover, a carefully designed iteratively reweighted training strategy is proposed to address the challenging multimodal inconsistency problem. We demonstrate the effectiveness of our approach on 7 computer vision tasks on videos. Extensive quantitative and perceptual experiments show that our approach obtains superior performance than state-of-the-art methods on blind video temporal consistency.

We present a fully automatic approach to video colorization with self-regularization and diversity. Our model contains a colorization network for video frame colorization and a refinement network for spatiotemporal color refinement. Without any labeled data, both networks can be trained with self-regularized losses defined in bilateral and temporal space. The bilateral loss enforces color consistency between neighboring pixels in a bilateral space and the temporal loss imposes constraints between corresponding pixels in two nearby frames. While video colorization is a multi-modal problem, our method uses a perceptual loss with diversity to differentiate various modes in the solution space. Perceptual experiments demonstrate that our approach outperforms state-of-the-art approaches on fully automatic video colorization.

A future video is the 2D projection of a 3D scene with predicted camera and object motion. Accurate future video prediction inherently requires understanding of 3D motion and geometry of a scene. In this paper, we propose a RGBD scene forecasting model with 3D motion decomposition. We predict ego-motion and foreground motion that are combined to generate a future 3D dynamic scene, which is then projected into a 2D image plane to synthesize future motion, RGB images and depth maps. Optional semantic maps can be integrated. Experimental results on KITTI and Driving datasets show that our model outperforms other state-of-the- arts in forecasting future RGBD dynamic scenes.

This paper shows that when applying machine learning to digital zoom, it is beneficial to operate on real, RAW sensor data. Existing learning-based super-resolution methods do not use real sensor data, instead operating on processed RGB images. We show that these approaches forfeit detail and accuracy that can be gained by operating on raw data, particularly when zooming in on distant objects. The key barrier to using real sensor data for training is that ground-truth high-resolution imagery is missing. We show how to obtain such ground-truth data via optical zoom and contribute a dataset, SR-RAW, for real-world computational zoom. We use SR-RAW to train a deep network with a novel contextual bilateral loss that is robust to mild misalignment between input and outputs images. The trained network achieves state-of-the-art performance in 4X and 8X computational zoom. We also show that synthesizing sensor data by resampling high-resolution RGB images is an oversimplified approximation of real sensor data and noise, resulting in worse image quality.

Deep learning has recently been applied with impressive results to extreme low-light imaging. Despite the success of single-image processing, extreme low-light video processing is still intractable due to the difficulty of collecting raw video data with corresponding ground truth. Collecting long-exposure ground truth, as was done for single-image processing, is not feasible for dynamic scenes. In this paper, we present deep processing of very dark raw videos: on the order of one lux of illuminance. To support this line of work, we collect a new dataset of raw low-light videos, in which high-resolution raw data is captured at video rate. At this level of darkness, the signal-to-noise ratio is extremely low (negative if measured in dB) and the traditional image processing pipeline generally breaks down. A new method is presented to address this challenging problem. By carefully designing a learning-based pipeline and introducing a new loss function to encourage temporal stability, we train a siamese network on static raw videos, for which ground truth is available, such that the network generalizes to videos of dynamic scenes at test time. Experimental results demonstrate that the presented approach outperforms state-of-the-art models for burst processing, per-frame processing, and blind temporal consistency.

We present a method for hiding video content inside audio files while preserving the perceptual fidelity of the cover audio. This is a form of cross-modal steganography and is particularly challenging due to the high bitrate of video. Our scheme uses recent advances in flow-based generative models, which enable mapping audio to latent codes such that nearby codes correspond to perceptually similar signals. We show that compressed video data can be concealed in the latent codes of audio sequences while preserving the fidelity of both the hidden video and the cover audio. We can embed 128x128 video inside same-duration audio, or higher-resolution video inside longer audio sequences. Quantitative experiments show that our approach outperforms relevant baselines in steganographic capacity and fidelity.

We present an end-to-end deep learning approach to denoising speech signals by processing the raw waveform directly. Given input audio containing speech corrupted by an additive background signal, the system aims to produce a processed signal that contains only the speech content. Recent approaches have shown promising results using various deep network architectures. In this paper, we propose to train a fully-convolutional context aggregation network using a deep feature loss. That loss is based on comparing the internal feature activations in a different network, trained for acoustic environment detection and domestic audio tagging. Our approach outperforms the state-of-the-art in objective speech quality metrics and in large-scale perceptual experiments with human listeners. It also outperforms an identical network trained using traditional regression losses. The advantage of the new approach is particularly pronounced for the hardest data with the most intrusive background noise, for which denoising is most needed and most challenging.

Imaging in low light is challenging due to low photon count and low SNR. Short-exposure images suffer from noise, while long exposure can induce blur and is often impractical. A variety of denoising, deblurring, and enhancement techniques have been proposed, but their effectiveness is limited in extreme conditions, such as video-rate imaging at night. To support the development of learning-based pipelines for low-light image processing, we introduce a dataset of raw short-exposure low-light images, with corresponding long-exposure reference images. Using the presented dataset, we develop a pipeline for processing low-light images, based on end-to-end training of a fully-convolutional network. The network operates directly on raw sensor data and replaces much of the traditional image processing pipeline, which tends to perform poorly on such data. We report promising results on the new dataset, analyze factors that affect performance, and highlight opportunities for future work.

Interactive image segmentation is characterized by multimodality. When the user clicks on a door, do they intend to select the door or the whole house? We present an end-to-end learning approach to interactive image segmentation that tackles this ambiguity. Our architecture couples two convolutional networks. The first is trained to synthesize a diverse set of plausible segmentations that conform to the user's input. The second is trained to select among these. By selecting a single solution, our approach retains compatibility with existing interactive segmentation interfaces. By synthesizing multiple diverse solutions before selecting one, the architecture is given the representational power to explore the multimodal solution space. We show that the proposed approach outperforms existing methods for interactive image segmentation, including prior work that applied convolutional networks to this problem, while being much faster.

We present a semi-parametric approach to photographic image synthesis from semantic layouts. The approach combines the complementary strengths of parametric and nonparametric techniques. The nonparametric component is a memory bank of image segments constructed from a training set of images. Given a novel semantic layout at test time, the memory bank is used to retrieve photographic references that are provided as source material to a deep network. The synthesis is performed by a deep network that draws on the provided photographic material. Experiments on multiple semantic segmentation datasets show that the presented approach yields considerably more realistic images than recent purely parametric techniques.

We present an approach to separating reflection from a single image. The approach uses a fully convolutional network trained end-to-end with losses that exploit low-level and high-level image information. Our loss function includes two perceptual losses: a feature loss from a visual perception network, and an adversarial loss that encodes characteristics of images in the transmission layers. We also propose a novel exclusion loss that enforces pixel-level layer separation. We create a dataset of real-world images with reflection and corresponding ground-truth transmission layers for quantitative evaluation and model training. We validate our method through comprehensive quantitative experiments and show that our approach outperforms state-of-the-art reflection removal methods in PSNR, SSIM, and perceptual user study. We also extend our method to two other image enhancement tasks to demonstrate the generality of our approach.

We present a learning-based approach to computing solutions for certain NP-hard problems. Our approach combines deep learning techniques with useful algorithmic elements from classic heuristics. The central component is a graph convolutional network that is trained to estimate the likelihood, for each vertex in a graph, of whether this vertex is part of the optimal solution. The network is designed and trained to synthesize a diverse set of solutions, which enables rapid exploration of the solution space via tree search. The presented approach is evaluated on four canonical NP-hard problems and five datasets, which include benchmark satisfiability problems and real social network graphs with up to a hundred thousand nodes. Experimental results demonstrate that the presented approach substantially outperforms recent deep learning work, and performs on par with highly optimized state-of-the-art heuristic solvers for some NP-hard problems. Experiments indicate that our approach generalizes across datasets, and scales to graphs that are orders of magnitude larger than those used during training.

We present an approach to synthesizing photographic images conditioned on semantic layouts. Given a semantic label map, our approach produces an image with photographic appearance that conforms to the input layout. The approach thus functions as a rendering engine that takes a two-dimensional semantic specification of the scene and produces a corresponding photographic image. Unlike recent and contemporaneous work, our approach does not rely on adversarial training. We show that photographic images can be synthesized from semantic layouts by a single feedforward network with appropriate structure, trained end-to-end with a direct regression objective. The presented approach scales seamlessly to high resolutions; we demonstrate this by synthesizing photographic images at 2-megapixel resolution, the full resolution of our training data. Extensive perceptual experiments on datasets of outdoor and indoor scenes demonstrate that images synthesized by the presented approach are considerably more realistic than alternative approaches.

We present an approach to accelerating a wide variety of image processing operators. Our approach uses a fully-convolutional network that is trained on input-output pairs that demonstrate the operator's action. After training, the original operator need not be run at all. The trained network operates at full resolution and runs in constant time. We investigate the effect of network architecture on approximation accuracy, runtime, and memory footprint, and identify a specific architecture that balances these considerations. We evaluate the presented approach on ten advanced image processing operators, including multiple variational models, multiscale tone and detail manipulation, photographic style transfer, nonlocal dehazing, and nonphoto-realistic stylization. All operators are approximated by the same model. Experiments demonstrate that the presented approach is significantly more accurate than prior approximation schemes. It increases approximation accuracy as measured by PSNR across the evaluated operators by 8.5 dB on the MIT-Adobe dataset (from 27.5 to 36 dB) and reduces DSSIM by a multiplicative factor of 3 compared to the most accurate prior approximation scheme, while being the fastest. We show that our models generalize across datasets and across resolutions, and investigate a number of extensions of the presented approach.

We present a global optimization approach to optical flow estimation. The approach optimizes a classical optical flow objective over the full space of mappings between discrete grids. No descriptor matching is used. The highly regular structure of the space of mappings enables optimizations that reduce the computational complexity of the algorithm's inner loop from quadratic to linear and support efficient matching of tens of thousands of nodes to tens of thousands of displacements. We show that one-shot global optimization of a classical Horn-Schunck-type objective over regular grids at a single resolution is sufficient to initialize continuous interpolation and achieve state-of-the-art performance on challenging modern benchmarks.

We present an approach to dense depth estimation from a single monocular camera that is moving through a dynamic scene. The approach produces a dense depth map from two consecutive frames. Moving objects are reconstructed along with the surrounding environment. We provide a novel motion segmentation algorithm that segments the optical flow field into a set of motion models, each with its own epipolar geometry. We then show that the scene can be reconstructed based on these motion models by optimizing a convex program. The optimization jointly reasons about the scales of different objects and assembles the scene in a common coordinate frame, determined up to a global scale. Experimental results demonstrate that the presented approach outperforms prior methods for monocular depth estimation in dynamic scenes.

We present an approach to nonrigid registration of 3D surfaces. We cast isometric embedding as MRF optimization and apply efficient global optimization algorithms based on linear programming relaxations. The Markov random field perspective suggests a natural connection with robust statistics and motivates robust forms of the intrinsic distortion functional. Our approach outperforms a large body of prior work by a significant margin, increasing registration precision on real data by a factor of 3.

We describe a simple and fast algorithm for optimizing Markov random fields over images. The algorithm performs block coordinate descent by optimally updating a horizontal or vertical line in each step. While the algorithm is not as accurate as state-of-the-art MRF solvers on traditional benchmark problems, it is trivially parallelizable and produces competitive results in a fraction of a second. As an application, we develop an approach to increasing the accuracy of consumer depth cameras. The presented algorithm enables high-resolution MRF optimization at multiple frames per second and substantially increases the accuracy of the produced range images.

We present a model for intrinsic decomposition of RGB-D images. Our approach analyzes a single RGB-D image and estimates albedo and shading fields that explain the input. To disambiguate the problem, our model estimates a number of components that jointly account for the reconstructed shading. By decomposing the shading field, we can build in assumptions about image formation that help distinguish reflectance variation from shading. These assumptions are expressed as simple nonlocal regularizers. We evaluate the model on real-world images and on a challenging synthetic dataset. The experimental results demonstrate that the presented approach outperforms prior models for intrinsic decomposition of RGB-D images.

This paper demonstrates how the nonlocal principle benefits video matting via the KNN Laplacian, which comes with a straightforward implementation using motion-aware K nearest neighbors. In hindsight, the fundamental problem to solve in video matting is to produce spatio-temporally coherent clusters of moving foreground pixels. When used as described, the motion-aware KNN Laplacian is effective in addressing this fundamental problem, as demonstrated by sparse user markups typically on only one frame in a variety of challenging examples featuring ambiguous foreground and background colors, changing topologies with disocclusion, significant illumination changes, fast motion, and motion blur. When working with existing Laplacian-based systems, we expect our Laplacian can benefit them immediately with an improved clustering of moving foreground pixels.

This paper proposes to apply the nonlocal principle to general alpha matting for the simultaneous extraction of multiple image layers; each layer may have disjoint as well as coherent segments typical of foreground mattes in natural image matting. The estimated alphas also satisfy the summation constraint. As in nonlocal matting, our approach does not assume the local color-line model and does not require sophisticated sampling or learning strategies. On the other hand, our matting method generalizes well to any color or feature space in any dimension, any number of alphas and layers at a pixel beyond two, and comes with an arguably simpler implementation, which we have made publicly available. Our matting technique, aptly called KNN matting, capitalizes on the nonlocal principle by using $(K)$ nearest neighbors (KNN) in matching nonlocal neighborhoods, and contributes a simple and fast algorithm that produces competitive results with sparse user markups. KNN matting has a closed-form solution that can leverage the preconditioned conjugate gradient method to produce an efficient implementation. Experimental evaluation on benchmark datasets indicates that our matting results are comparable to or of higher quality than state-of-the-art methods requiring more involved implementation. In this paper, we take the nonlocal principle beyond alpha estimation and extract overlapping image layers using the same Laplacian framework. Given the alpha value, our closed form solution can be elegantly generalized to solve the multilayer extraction problem. We perform qualitative and quantitative comparisons to demonstrate the accuracy of the extracted image layers.

We are interested in a general alpha matting approach for the simultaneous extraction of multiple image layers; each layer may have disjoint segments for material matting not limited to foreground mattes typical of natural image matting. The estimated alphas also satisfy the summation constraint. Our approach does not assume the local color-line model, does not need sophisticated sampling strategies, and generalizes well to any color or feature space in any dimensions. Our matting technique, aptly called KNN matting, capitalizes on the nonlocal principle by using K nearest neighbors (KNN) in matching nonlocal neighborhoods, and contributes a simple and fast algorithm giving competitive results with sparse user markups. KNN matting has a closed-form solution that can leverage on the preconditioned conjugate gradient method to produce an efficient implementation. Experimental evaluation on benchmark datasets indicates that our matting results are comparable to or of higher quality than state of the art methods.