Exploring databases remains cognitively demanding for non-experts. While natural language interfaces offer flexibility, they place the full burden of articulation and refinement on the user, hindering exploratory discovery. We identify a core interaction design problem: how to dynamically support users’ evolving understanding during query formulation. We propose and implement a paradigm of adaptive mixed-initiative interaction, where the system interprets user behavioral cues (e.g., tentativeness, focus shifts) to infer intent and dynamically adapts its support strategies. This involves switching between responsive and proactive modes, and integrating textual responses with graphical previews to scaffold the query-building process. We instantiate this paradigm in a functional prototype. A controlled user study demonstrates that this adaptive approach not only improves task efficiency and usability but, more importantly, reduces cognitive load and fosters a more exploratory, less formulaic query-building process compared to traditional reactive interfaces.

Although Speech Large Language Models have achieved notable progress, a substantial modality reasoning gap remains: their reasoning performance on speech inputs is markedly weaker than on text.This gap could be associated with representational drift across Transformer layers and behavior deviations in long-chain reasoning.To address this issue, we introduce TARS, a reinforcement-learning framework that aligns text-conditioned and speech-conditioned trajectories through an asymmetric reward design.The framework employs two dense and complementary signals: representation alignment, which measures layer-wise hidden-state similarity between speech-and text-conditioned trajectories, and behavior alignment, which evaluates semantic consistency between generated outputs and reference text completions.Experiments on challenging reasoning benchmarks, including MMSU and OBQA, show that our approach significantly narrows the modality reasoning gap and achieves state-of-the-art performance among 7B-scale Speech LLMs.

Hui Wang, Jinghua Zhao, Yifan Yang, Shujie Liu, Junyang Chen, Yanzhe Zhang, Shiwan Zhao, Jinyu Li, Jiaming Zhou, Haoqin Sun, Yan Lu, Yong Qin. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Tables contain rich structured information, yet when stored as images their contents remain "locked" within pixels.Converting table images into LaTeX code enables faithful digitization and reuse, but current multimodal large language models (MLLMs) often fail to preserve structural, style, or content fidelity.Conventional post-training with reinforcement learning (RL) typically relies on a single aggregated reward, leading to reward ambiguity that conflates multiple behavioral aspects and hinders effective optimization.We propose Component-Specific Policy Optimization (CSPO), an RL framework that disentangles optimization across LaTeX tables components-structure, style, and content.In particular, CSPO assigns component-specific rewards and backpropagates each signal only through the tokens relevant to its component, alleviating reward ambiguity and enabling targeted component-wise optimization.To comprehensively assess performance, we introduce a set of hierarchical evaluation metrics.Extensive experiments demonstrate the effectiveness of CSPO, underscoring the importance of component-specific optimization for reliable structured generation.

GUI agents that interact with graphical interfaces on behalf of users are a promising direction for practical AI assistants, yet training them is hindered by scarce suitable environments. We present InfiniteWeb, a system that automatically generates functional web environments at scale for GUI agent training. While LLMs perform well on generating a single webpage, building a realistic and functional website with many interconnected pages faces challenges. We address these challenges through unified specification, task-centric test-driven development, and combining website seed variation with reference design images. Our system also generates verifiable task evaluators enabling dense reward signals for reinforcement learning. Experiments show that our system surpasses commercial coding agents at realistic website construction, and GUI agents trained on our generated environments achieve significant performance improvements on OSWorld and Online-Mind2Web, demonstrating the effectiveness of the proposed system.

Recent advances in foundation models have enabled conversational agents that aim for sustained companionship rather than mere task completion. Yet most still remain unable to support natural, long-term companion-like interactions, resulting in experiences that feel episodic and inauthentic. We argue that current agents overlooked cross-temporal modeling of agents’ social behaviors and internal emotions: generated behaviors rarely influence an agent’s emotional state, and emotional states seldom shape subsequent behaviors. We present Cross-Temporal Emotion Modeling (CTEM), a framework that links long-term behavioral history to moment-to-moment emotional expression. CTEM establishes a closed loop where past experiences update an evolving emotional state; this state conditions immediate interactions; and user feedback continually revises both memory and emotional state, enabling reflection and anticipation. We instantiate CTEM as Auri, a companion agent on an instant-messaging platform, and report a 21-day in-the-wild study showing that CTEM shows improvements in perceived naturalness, coherence, and emotional harmony.

Video object removal aims at erasing a target object in the entire video and filling holes with plausible contents, given an object mask in the first frame as input. Existing solutions mostly break down the task into (supervised) mask tracking and (self-supervised) video completion, and then separately tackle them with tailored designs. In this paper, we introduce a new setup, coined as unified video object removal, where mask tracking and completion are addressed within a unified framework. Despite introducing more challenges, the setup is promising for future practical usage. We embrace the observation that these two sub-tasks have strong inherent connections in terms of pixel-level temporal correspondence. Making full use of the connections could be beneficial considering the complexity of both algorithm and deployment. We propose a single network linking the two sub-tasks by inferring temporal correspondences across multiple frames, i.e., correspondences between valid-valid (V-V) pixel pairs for mask tracking and correspondences between valid-hole (V-H) pixel pairs for video completion. Thanks to the unified setup, the network can be learned end-to-end in a totally unsupervised fashion without any annotations. We demonstrate that our method can generate visually pleasing results and perform favorably against existing separate solutions in realistic test cases.

The remarkable natural language understanding, reasoning, and generation capabilities of large language models (LLMs) have made them attractive for application to video understanding, utilizing video tokens as contextual input. However, employing LLMs for long video understanding presents significant challenges. The extensive number of video tokens leads to considerable computational costs for LLMs while using aggregated tokens results in loss of vision details. Moreover, the presence of abundant question-irrelevant tokens introduces noise to the video reasoning process. To address these issues, we introduce a simple yet effective learnable retrieval-based video-language model (R-VLM) for efficient long video understanding. Specifically, given a question and a long video, our model identifies the most relevant$K$video chunks and uses their associated visual tokens to serve as context for the LLM inference. This effectively reduces the number of video tokens, eliminates noise interference, and enhances system performance. We achieve this by incorporating a learnable lightweight MLP block to facilitate the efficient retrieval of question-relevant chunks, through the end-to-end training of our video-language model with a proposed soft matching loss. Experimental results on multiple zero-shot video question answering datasets validate the effectiveness of our framework for comprehending long videos.

Vision-language temporal alignment is a crucial capability for human dynamic recognition and cognition in real-world scenarios. While existing research focuses on capturing vision-language relevance, it faces limitations due to biased temporal distributions, imprecise annotations, and insufficient compositionally. To achieve fair evaluation and comprehensive exploration, our objective is to investigate and evaluate the ability of models to achieve alignment from a temporal perspective, specifically focusing on their capacity to synchronize visual scenarios with linguistic context in a temporally coherent manner. As a preliminary step, we present the statistical analysis of existing benchmarks and reveal the existing challenges from a decomposed perspective. To this end, we introduce SVLTA, the Synthetic Vision-Language Temporal Alignment derived via a well-designed and feasible control generation method within a simulation environment. The approach considers commonsense knowledge, manipulable action, and constrained filtering, which generates reasonable, diverse, and balanced data distributions for diagnostic evaluations. Our experiments reveal diagnostic insights through the evaluations in temporal question answering, distributional shift sensitiveness, and temporal alignment adaptation.

Recent advances in image-to-video generation have enabled animation of still images and offered pixel-level controllability. While these models hold great potential to transform single images into vivid and dynamic videos, they also carry risks of misuse that could impact privacy, security, and copyright protection. This paper proposes a novel approach that applies imperceptible perturbations on images to degrade the quality of the generated videos, thereby protecting images from misuse in white-box image-to-video diffusion models. Specifically, we function our approach as an adversarial attack, incorporating spatial, temporal, and diffusion attack modules. The spatial attack shifts image features from their original distribution to a lower-quality target distribution, reducing visual fidelity. The temporal attack disrupts coherent motion by interfering with temporal attention maps that guide motion generation. To enhance the robustness of our approach across different models, we further propose a diffusion attack module leveraging contrastive loss. Our approach can be easily integrated with mainstream diffusion-based I2V models. Extensive experiments on SVD, CogVideoX, and ControlNeXt demonstrate that our method significantly impairs generation quality in terms of visual clarity and motion consistency, while introducing only minimal artifacts to the images. To the best of our knowledge, we are the first to explore adversarial attacks on image-to-video generation for security purposes.

We introduce a practical real-time neural video codec (NVC) designed to deliver high compression ratio, low latency and broad versatility. In practice, the coding speed of NVCs depends on 1) computational costs, and 2) non-computational operational costs, such as memory I/O and the number of function calls. While most efficient NVCs prioritize reducing computational cost, we identify operational cost as the primary bottleneck to achieving higher coding speed. Leveraging this insight, we introduce a set of efficiency-driven design improvements focused on minimizing operational costs. Specifically, we employ implicit temporal modeling to eliminate complex explicit motion modules, and use single low-resolution latent representations rather than progressive downsampling. These innovations significantly accelerate NVC without sacrificing compression quality. Additionally, we implement model integerization for consistent cross-device coding and a module-bank-based rate control scheme to improve practical adaptability. Experiments show our proposed DCVC-RT achieves an impressive average encoding/decoding speed at 125.2/112.8 fps (frames per second) for 1080p video, while saving an average of 21% in bitrate compared to H.266/VTM. The code is available at https://github.com/microsoft/DCVC.

We introduce UniGraspTransformer, a universal Transformer-based network for dexterous robotic grasping that simplifies training while enhancing scalability and performance. Unlike prior methods such as UniDexGrasp++, which require complex, multi-step training pipelines, UniGraspTransformer follows a streamlined process: first, dedicated policy networks are trained for individual objects using reinforcement learning to generate successful grasp trajectories; then, these trajectories are distilled into a single, universal network. Our approach enables UniGraspTransformer to scale effectively, incorporating up to 12 self-attention blocks for handling thousands of objects with diverse poses. Additionally, it generalizes well to both idealized and real-world inputs, evaluated in state-based and vision-based settings. Notably, UniGraspTransformer generates a broader range of grasping poses for objects in various shapes and orientations, resulting in more diverse grasp strategies. Experimental results demonstrate significant improvements over state-of-the-art, UniDexGrasp++, across various object categories, achieving success rate gains of 3.5%, 7.7%, and 10.1% on seen objects, unseen objects within seen categories, and completely unseen objects, respectively, in the vision-based setting. Project page: https://dexhand.github.io/UniGraspTransformer/.

Recently, perceptual image compression has achieved significant advancements, delivering high visual quality at low bitrates for natural images. However, for screen content, existing methods often produce noticeable artifacts when compressing text. To tackle this challenge, we propose versatile perceptual screen image compression with diffusion rendering (PICD), a codec that works well for both screen and natural images. More specifically, we propose a compression framework that encodes the text and image separately, and renders them into one image using diffusion model. For this diffusion rendering, we integrate conditional information into diffusion models at three distinct levels: 1). Domain level: We fine-tune the base diffusion model using text content prompts with screen content. 2). Adaptor level: We develop an efficient adaptor to control the diffusion model using compressed image and text as input. 3). Instance level: We apply instance-wise guidance to further enhance the decoding process. Empirically, our PICD surpasses existing perceptual codecs in terms of both text accuracy and perceptual quality. Additionally, without text conditions, our approach serves effectively as a perceptual codec for natural images.

Zezhong Jin, Shubhang Desai, Xu Chen, Biyi Fang, Zhuoyi Huang, Zhe Li, Chong-Xin Gan, Xiao Tu, Man-Wai Mak, Yan Lu, Shujie Liu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

We propose a novel and general framework to disentangle video data into its dynamic motion and static content components. Our proposed method is a self-supervised pipeline with less assumptions and inductive biases than previous works: it utilizes a transformer-based architecture to jointly generate flexible implicit features for frame-wise motion and clip-wise content, and incorporates a low-bitrate vector quantization as an information bottleneck to promote disentanglement and form a meaningful discrete motion space. The bitrate-controlled latent motion and content are used as conditional inputs to a denoising diffusion model to facilitate self-supervised representation learning. We validate our disentangled representation learning framework on real-world talking head videos with motion transfer and auto-regressive motion generation tasks. Furthermore, we also show that our method can generalize to other types of video data, such as pixel sprites of 2D cartoon characters. Our work presents a new perspective on self-supervised learning of disentangled video representations, contributing to the broader field of video analysis and generation.

In real applications, person re-identification (ReID) expects to retrieve the target person at any time, including both daytime and nighttime, ranging from short-term to long-term. However, existing ReID tasks and datasets cannot meet this requirement, as they are constrained by available time and only provide training and evaluation for specific scenarios. Therefore, we investigate a new task called Anytime Person Re-identification (AT-ReID), which aims to achieve effective retrieval in multiple scenarios based on variations in time. To address the AT-ReID problem, we collect the first large-scale dataset, AT-USTC, which contains 135k images of individuals wearing multiple clothes captured by RGB and IR cameras. Our data collection spans over an entire year and 270 volunteers were photographed on average 29.1 times across different dates or scenes, 4-15 times more than current datasets, providing conditions for follow-up investigations in AT-ReID. Further, to tackle the new challenge of multi-scenario retrieval, we propose a unified model named Uni-AT, which comprises a multi-scenario ReID (MS-ReID) framework for scenario-specific features learning, a Mixture-of-Attribute-Experts (MoAE) module to alleviate inter-scenario interference, and a Hierarchical Dynamic Weighting (HDW) strategy to ensure balanced training across all scenarios. Extensive experiments show that our model leads to satisfactory results and exhibits excellent generalization to all scenarios.

Recent zero-shot text-to-speech (TTS) systems face a common dilemma: autoregressive (AR) models suffer from slow generation and lack duration controllability, while non-autoregressive (NAR) models lack temporal modeling and typically require complex designs. In this paper, we introduce a novel pseudo-autoregressive (PAR) codec language modeling approach that unifies AR and NAR modeling. Combining explicit temporal modeling from AR with parallel generation from NAR, PAR generates dynamic-length spans at fixed time steps. Building on PAR, we propose PALLE, a two-stage TTS system that leverages PAR for initial generation followed by NAR refinement. In the first stage, PAR progressively generates speech tokens along the time dimension, with each step predicting all positions in parallel but only retaining the left-most span. In the second stage, low-confidence tokens are iteratively refined in parallel, leveraging the global contextual information. Experiments demonstrate that PALLE, trained on LibriTTS, outperforms state-of-the-art systems trained on large-scale data, including F5-TTS, E2-TTS, and MaskGCT, on the LibriSpeech test-clean set in terms of speech quality, speaker similarity, and intelligibility, while achieving up to ten times faster inference speed. Audio samples are available at https://microsoft.com/research/project/vall-e-x/palle.

To advance continuous token modeling and temporal-coherence enforcement, we propose FELLE, an autoregressive model that integrates language modeling with token-wise flow matching. By leveraging the autoregressive nature of language models and the generative efficacy of flow matching, FELLE effectively predicts continuous-valued tokens (mel-spectrograms). For each continuous-valued token, FELLE modifies the general prior distribution in flow matching by incorporating information from the previous step, improving coherence and stability. Furthermore, to enhance synthesis quality, FELLE introduces a coarse-to-fine flow-matching mechanism, generating continuous-valued tokens hierarchically, conditioned on the language model's output. Experimental results demonstrate the potential of incorporating flow-matching techniques in autoregressive mel-spectrogram modeling, leading to significant improvements in TTS generation quality, as shown in https://aka.ms/felle.

The burgeoning growth of open-source vision-language models (VLMs) has catalyzed a plethora of applications across diverse domains. Ensuring the transparency and interpretability of these models is critical for fostering trustworthy and responsible AI systems. In this study, our objective is to delve into the internals of VLMs to interpret the functions of individual neurons. We observe the activations of neurons with respects to the input visual tokens and text tokens, and reveal some interesting findings. Particularly, we found that there are neurons responsible for only visual or text information, or both, respectively, which we refer to them as visual neurons, text neurons, and multi-modal neurons, respectively. We build a framework that automates the explanation of neurons with the assistant of GPT-4o. Meanwhile, for visual neurons, we propose an activation simulator to assess the reliability of the explanations for visual neurons. System statistical analyses on top of one representative VLM of LLaVA, uncover the behaviors/characteristics of different categories of neurons.

Generating an interactive visual world from a single image is both challenging and practically valuable, as single-view inputs are easy to acquire and align well with prompt-driven applications such as gaming and virtual reality. This paper introduces a novel unified framework, Image as a World (**IaaW**), which synthesizes high-quality 360-degree videos from a single image that are both controllable and temporally continuable. Our framework consists of three stages: world initialization, which jointly synthesizes spatially complete and temporally dynamic scenes from a single view; world exploration, which supports user-specified viewpoint rotation; and world continuation, which extends the generated scene forward in time with temporal consistency. To support this pipeline, we design a visual world model based on generative diffusion models modulated with spherical 3D positional encoding and multi-view composition to represent geometry and view semantics. Additionally, a vision-language model (IaaW-VLM) is fine-tuned to produce both global and view-specific prompts, improving semantic alignment and controllability. Extensive experiments demonstrate that our method produces panoramic videos with superior visual quality, minimal distortion and seamless continuation in both qualitative and quantitative evaluations. To the best of our knowledge, this is the first work to generate a controllable, consistent, and temporally expandable 360-degree world from a single image.

Generalization in robot manipulation is essential for deploying robots in open-world environments and advancing toward artificial general intelligence. While recent Vision-Language-Action (VLA) models leverage large pre-trained understanding models for perception and instruction following, their ability to generalize to novel tasks, objects, and settings remains limited. In this work, we present VideoVLA, a simple approach that explores the potential of transforming large video generation models into robotic VLA manipulators. Given a language instruction and an image, VideoVLA predicts an action sequence as well as the future visual outcomes. Built on a multi-modal Diffusion Transformer, VideoVLA jointly models video, language, and action modalities, using pre-trained video generative models for joint visual and action forecasting. Our experiments show that high-quality imagined futures correlate with reliable action predictions and task success, highlighting the importance of visual imagination in manipulation. VideoVLA demonstrates strong generalization, including imitating other embodiments' skills and handling novel objects. This dual-prediction strategy—forecasting both actions and their visual consequences—explores a paradigm shift in robot learning and unlocks generalization capabilities in manipulation systems.

Super-resolution (SR) advances astronomical imaging by enabling cost-effective high-resolution capture, crucial for detecting faraway celestial objects and precise structural analysis. However, existing datasets for astronomical SR (ASR) exhibit three critical limitations: flux inconsistency, object-crop setting, and insufficient data diversity, significantly impeding ASR development. We propose STAR, a large-scale astronomical SR dataset containing 54,738 flux-consistent star field image pairs covering wide celestial regions. These pairs combine Hubble Space Telescope high-resolution observations with physically faithful low-resolution counterparts generated through a flux-preserving data generation pipeline, enabling systematic development of field-level ASR models. To further empower the ASR community, STAR provides a novel Flux Error (FE) to evaluate SR models in physical view. Leveraging this benchmark, we propose a Flux-Invariant Super Resolution (FISR) model that could accurately infer the flux-consistent high-resolution images from input photometry, suppressing several SR state-of-the-art methods by 24.84% on a novel designed flux consistency metric, showing the priority of our method for astrophysics. Extensive experiments demonstrate the effectiveness of our proposed method and the value of our dataset. Code and models are available at https://github.com/GuoCheng12/STAR.

While recent diffusion-based generative image codecs have shown impressive performance, their iterative sampling process introduces unpleasant latency. In this work, we revisit the design of a diffusion-based codec and argue that multi-step sampling is not necessary for generative compression. Based on this insight, we propose OneDC, a One-step Diffusion-based generative image Codec—that integrates a latent compression module with a one-step diffusion generator. Recognizing the critical role of semantic guidance in one-step diffusion, we propose using the hyperprior as a semantic signal, overcoming the limitations of text prompts in representing complex visual content. To further enhance the semantic capability of the hyperprior, we introduce a semantic distillation mechanism that transfers knowledge from a pretrained generative tokenizer to the hyperprior codec. Additionally, we adopt a hybrid pixel- and latent-domain optimization to jointly enhance both reconstruction fidelity and perceptual realism. Extensive experiments demonstrate that OneDC achieves SOTA perceptual quality even with one-step generation, offering over 39% bitrate reduction and 20× faster decoding compared to prior multi-step diffusion-based codecs. Project: https://onedc-codec.github.io/

Reconstruction of 3D scenes from a single image is a crucial step towards enabling next-generation AI-powered immersive experiences. However, existing diffusion-based methods often struggle with reconstructing omnidirectional scenes due to geometric distortions and inconsistencies across the generated novel views, hindering accurate 3D recovery. To overcome this challenge, we propose Omni3D, an approach designed to enhance the geometric fidelity of diffusion-generated views for robust omnidirectional reconstruction. Our method leverages priors from pose estimation techniques, such as MASt3R, to iteratively refine both the generated novel views and their estimated camera poses. Specifically, we minimize the 3D reprojection errors between paired views to optimize the generated images, and simultaneously, correct the pose estimation based on the refined views. This synergistic optimization process yields geometrically consistent views and accurate poses, which are then used to build an explicit 3D Gaussian Splatting representation capable of omnidirectional rendering. Experimental results validate the effectiveness of Omni3D, demonstrating significantly advanced 3D reconstruction quality in the omnidirectional space, compared to previous state-of-the-art methods. Project page: https://omni3d-neurips.github.io.

Long-form video understanding presents significant challenges due to extensive temporal-spatial complexity and the difficulty of question answering under such extended contexts. While Large Language Models (LLMs) have demonstrated considerable advancements in video analysis capabilities and long context handling, they continue to exhibit limitations when processing information-dense hour-long videos. To overcome such limitations, we propose the $\textbf{D}eep \ \textbf{V}ideo \ \textbf{D}iscovery \ (\textbf{DVD})$ agent to leverage an $\textit{agentic search}$ strategy over segmented video clips. Different from previous video agents manually designing a rigid workflow, our approach emphasizes the autonomous nature of agents. By providing a set of search-centric tools on multi-granular video database, our DVD agent leverages the advanced reasoning capability of LLM to plan on its current observation state, strategically selects tools to orchestrate adaptive workflow for different queries in light of the gathered information. We perform comprehensive evaluation on multiple long video understanding benchmarks that demonstrates our advantage. Our DVD agent achieves state-of-the-art performance on the challenging LVBench dataset, reaching an accuracy of $\textbf{74.2\%}$, which substantially surpasses all prior works, and further improves to $\textbf{76.0\%}$ with transcripts.

Aircraft manufacturing is the jewel in the crown of industry, in which generating high-fidelity airfoil geometries with controllable and editable representations remains a fundamental challenge. Existing deep learning methods, which typically rely on predefined parametric representations (e.g., Bézier curves) or discrete point sets, face an inherent trade-off between expressive power and resolution adaptability. To tackle this challenge, we introduce FuncGenFoil, a novel function-space generative model that directly reconstructs airfoil geometries as function curves. Our method inherits the advantages of arbitrary-resolution sampling and smoothness from parametric functions, as well as the strong expressiveness of discrete point-based representations. Empirical evaluations demonstrate that FuncGenFoil improves upon state-of-the-art methods in airfoil generation, achieving a relative 74.4% reduction in label error and a 23.2% increase in diversity on the AF-200K dataset. Our results highlight the advantages of function-space modeling for aerodynamic shape optimization, offering a powerful and flexible framework for high-fidelity airfoil design.

Deep learning-based computational methods have achieved promising results in predicting protein-protein interactions (PPIs). However, existing benchmarks predominantly focus on isolated pairwise evaluations, overlooking a model's capability to reconstruct biologically meaningful PPI networks, which is crucial for biology research. To address this gap, we introduce PRING, the first comprehensive benchmark that evaluates PRotein-protein INteraction prediction from a Graph-level perspective. PRING curates a high-quality, multi-species PPI network dataset comprising 21,484 proteins and 186,818 interactions, with well-designed strategies to address both data redundancy and leakage. Building on this golden-standard dataset, we establish two complementary evaluation paradigms: (1) topology-oriented tasks, which assess intra and cross-species PPI network construction, and (2) function-oriented tasks, including protein complex pathway prediction, GO module analysis, and essential protein justification. These evaluations not only reflect the model's capability to understand the network topology but also facilitate protein function annotation, biological module detection, and even disease mechanism analysis. Extensive experiments on four representative model categories, consisting of sequence similarity-based, naive sequence-based, protein language model-based, and structure-based approaches, demonstrate that current PPI models have potential limitations in recovering both structural and functional properties of PPI networks, highlighting the gap in supporting real-world biological applications. We believe PRING provides a reliable platform to guide the development of more effective PPI prediction models for the community. The dataset and source code of PRING are available at https://github.com/SophieSarceau/PRING.

Scientific discoveries increasingly rely on complex multimodal reasoning based on information-intensive scientific data and domain-specific expertise. Empowered by expert-level scientific benchmarks, scientific Multimodal Large Language Models (MLLMs) hold the potential to significantly enhance this discovery process in realistic workflows. However, current scientific benchmarks mostly focus on evaluating the knowledge understanding capabilities of MLLMs, leading to an inadequate assessment of their perception and reasoning abilities. To address this gap, we present the Scientists’ First Exam (SFE) benchmark, designed to evaluate the scientific cognitive capacities of MLLMs through three interconnected levels: scientific signal perception, scientific attribute understanding, scientific comparative reasoning. Specifically, SFE comprises 830 expert-verified VQA pairs across three question types, spanning 66 multimodal tasks across five high-value disciplines. Extensive experiments reveal that current state-of-the-art GPT-o3 and InternVL-3 achieve only 34.08% and 26.52% on SFE, highlighting significant room for MLLMs to improve in scientific realms. We hope the insights obtained in SFE will facilitate further developments in AI-enhanced scientific discoveries.

Geometry plays a significant role in monocular 3D object detection. It can be used to estimate object depth by using the perspective projection between object's physical size and 2D projection in the image plane, which can introduce mathematical priors into deep models. However, this projection process also introduces error amplification, where the error of the estimated height is amplified and reflected into the projected depth. It leads to unreliable depth inferences and also impairs training stability. To tackle this problem, we propose a novel Geometry Uncertainty Propagation Network (GUPNet++) by modeling geometry projection in a probabilistic manner. This ensures depth predictions are well-bounded and associated with a reasonable uncertainty. The significance of introducing such geometric uncertainty is two-fold: (1). It models the uncertainty propagation relationship of the geometry projection during training, improving the stability and efficiency of the end-to-end model learning. (2). It can be derived to a highly reliable confidence to indicate the quality of the 3D detection result, enabling more reliable detection inference. Experiments show that the proposed approach not only obtains (state-of-the-art) SOTA performance in image-based monocular 3D detection but also demonstrates superiority in efficacy with a simplified framework. The code and model will be released at https://github.com/SuperMHP/GUPNet_Plus.

Recent advances in zero-shot text-to-speech (TTS) synthesis have achieved high-quality speech generation for unseen speakers, but most systems remain unsuitable for real-time applications because of their offline design. Current streaming TTS paradigms often rely on multi-stage pipelines and discrete representations, leading to increased computational cost and suboptimal system performance. In this work, we propose StreamMel, a pioneering single-stage streaming TTS framework that models continuous mel-spectrograms. By interleaving text tokens with acoustic frames, StreamMel enables low-latency, autoregressive synthesis while preserving high speaker similarity and naturalness. Experiments on LibriSpeech demonstrate that StreamMel outperforms existing streaming TTS baselines in both quality and latency. It even achieves performance comparable to offline systems while supporting efficient real-time generation, showcasing broad prospects for integration with real-time speech large language models. Audio samples are available at:https://aka.ms/StreamMel.

Because of the ambiguous and subjective property of the facial expression, the label noise is widely existing in the FER dataset. For this problem, in the training phase, current methods often directly predict whether the label is noised or not, aiming to reduce the contribution of the noised data. However, we argue that this kind of method suffers from the low reliability of such noise data decision operation. It makes that some mistakenly abounded clean data are not utilized sufficiently and some mistakenly kept noised data disturbing the model learning. In this paper, we propose a more reliable noise-label suppression method called ReSup. First, instead of directly predicting noised or not, ReSup makes the noise data decision by modeling the distribution of noise and clean labels simultaneously according to the disagreement between the prediction and the target. Specifically, to achieve optimal distribution modeling, ReSup models the similarity distribution of all samples. To further enhance the reliability of our noise decision results, ReSup uses two networks to jointly achieve noise suppression. Specifically, ReSup utilize the property that two networks are less likely to make the same mistakes, making two networks swap decisions and tending to trust decisions with high agreement. Extensive experiments on popular datasets shows the effectiveness of ReSup.

Most existing approaches for image and video compression perform transform coding in the pixel space to reduce redundancy. However, due to the misalignment between the pixel-space distortion and human perception, such schemes often face the difficulties in achieving both high-realism and high-fidelity at ultra-low bitrate. To solve this problem, we propose Generative Latent Coding (GLC) models for image and video compression, termed GLC-image and GLC-Video. The transform coding of GLC is conducted in the latent space of a generative vector-quantized variational auto-encoder (VQ-VAE). Compared to the pixel-space, such a latent space offers greater sparsity, richer semantics and better alignment with human perception, and show its advantages in achieving high-realism and high-fidelity compression. To further enhance performance, we improve the hyper prior by introducing a spatial categorical hyper module in GLC-image and a spatio-temporal categorical hyper module in GLC-video. Additionally, the code-prediction-based loss function is proposed to enhance the semantic consistency. Experiments demonstrate that our scheme shows high visual quality at ultra-low bitrate for both image and video compression. For image compression, GLC-image achieves an impressive bitrate of less than 0.04 bpp, achieving the same FID as previous SOTA model MS-ILLM while using 45% fewer bitrate on the CLIC 2020 test set. For video compression, GLC-video achieves 65.3% bitrate saving over PLVC in terms of DISTS.

As advancements are made in technology such as AR/VR and high-resolution photography, there is a growing need for a function in image compression named regional decoding . This function lets an image be encoded as a whole, but allows for an arbitrary region to be decoded using only a small part of the bitstream. However, existing neural image compression methods lack support for this crucial functionality. In this article, we propose a novel approach called the slicing en/decoder , which addresses the need for regional decoding while maintaining performance on par with state-of-the-art methods. Our approach is based on the insight that, during the compression process, local information within pixels holds greater importance than global information. By leveraging this understanding, we divide the image into different bitstreams according to cross-boundary patterns. Consequently, for a selected region, our method can intelligently choose specific portions of the bitstreams to decode only that particular region of interest. Furthermore, we extend the application of our method to 360° image compression, allowing for efficient encoding and decoding of immersive visual content. Moreover, our proposed technique offers the capability to decode regions identically, which paves the way for future advancements in regional video decoding. Our experimental results demonstrate that our method maintains performance on par with state-of-the-art methods while providing the functionality of regional decoding . In conclusion, this article presents a significant step forward in image compression technology, offering enhanced flexibility and efficiency for emerging applications in digital media.

We propose a Dynamic Context-Guided Upsampling (DCGU) module for video super-resolution (VSR) that leverages temporal context guidance to achieve efficient and effective arbitrary-scale VSR. While most VSR research focuses on backbone design, the importance of the upsampling part is often overlooked. Existing methods rely on pixelshuffle-based upsampling, which has limited capabilities in handling arbitrary upsampling scales. Recent attempts to replace pixelshuffle-based modules with implicit neural function-based and filter-based approaches suffer from slow inference speeds and limited representation capacity, respectively. To overcome these limitations, our DCGU module predicts non-local sampling locations and content-dependent filter weights, enabling efficient and effective arbitrary-scale VSR. Our proposed multi-granularity location search module efficiently identifies non-local sampling locations across the entire low-resolution grid, and the temporal bilateral filter modulation module integrates content information with the filter weight to enhance textual details. Extensive experiments demonstrate the superiority of our method in terms of performance and speed on arbitrary-scale VSR.

Generating realistic human motion from given action descriptions has experienced significant advancements because of the emerging requirement of digital humans. While recent works have achieved impressive results in generating motion directly from textual action descriptions, they often support only a single modality of the control signal, which limits their application in the real digital human industry. This paper presents a Motion General-Purpose generaTor (MotionGPT) that can use multimodal control signals, e.g., text and single-frame poses, for generating consecutive human motions by treating multimodal signals as special input tokens in large language models (LLMs). Specifically, we first quantize multimodal control signals into discrete codes and then formulate them in a unified prompt instruction to ask the LLMs to generate the motion answer. Our MotionGPT demonstrates a unified human motion generation model with multimodal control signals by tuning a mere 0.4% of LLM parameters. To the best of our knowledge, MotionGPT is the first method to generate human motion by multimodal control signals, which we hope can shed light on this new direction. Visit our webpage at https://qiqiapink.github.io/MotionGPT/.

Text-to-Image person re-identification (TI-ReID) aims to retrieve the images of target identity according to the given textual description. The existing methods in TI-ReID focus on aligning the visual and textual modalities through contrastive feature alignment or reconstructive masked language modeling (MLM). However, these methods parameterize the image/text instances as deterministic embeddings and do not explicitly consider the inherent uncertainty in pedestrian images and their textual descriptions, leading to limited image-text relationship expression and semantic alignment. To address the above problem, in this paper, we propose a novel method that unifies multi-modal uncertainty modeling and semantic alignment for TI-ReID. Specifically, we model the image and textual feature vectors of pedestrian as Gaussian distributions, where the multi-granularity uncertainty of the distribution is estimated by incorporating batch-level and identity-level feature variances for each modality. The multi-modal uncertainty modeling acts as a feature augmentation and provides richer image-text semantic relationship. Then we present a bi-directional cross-modal circle loss to more effectively align the probabilistic features between image and text in a self-paced manner. To further promote more comprehensive image-text semantic alignment, we design a task that complements the masked language modeling, focusing on the cross-modality semantic recovery of global masked token after cross-modal interaction. Extensive experiments conducted on three TI-ReID datasets highlight the effectiveness and superiority of our method over state-of-the-arts.

Audiovisual segmentation (AVS) is a challenging task that aims to segment visual objects in videos according to their associated acoustic cues. With multiple sound sources and background disturbances involved, establishing robust correspondences between audio and visual contents poses unique challenges due to (1) complex entanglement across sound sources and (2) frequent changes in the occurrence of distinct sound events. Assuming sound events occur in- dependently, the multi-source semantic space can be rep- resented as the Cartesian product of single-source sub- spaces. We are motivated to decompose the multi-source audio semantics into single-source semantics for more ef- fective interactions with visual content. We propose a se- mantic decomposition method based on product quanti- zation, where the multi-source semantics can be decom- posed and represented by several disentangled and noise- suppressed single-source semantics. Furthermore, we in- troduce a global-to-local quantization mechanism, which distills knowledge from stable global (clip-level) features into local (frame-level) ones, to handle frequent changes in audio semantics. Extensive experiments demonstrate that our semantically decomposed audio representation signifi- cantly improves AVS performance, e.g., +21.2% mIoU on the challenging AVS-Semantic benchmark with ResNet50 backbone.

Significant progress has been made in scene text detection models since the rise of deep learning, but scene text layout analysis, which aims to group detected text instances as paragraphs, has not kept pace. Previous works either treated text detection and grouping using separate models, or train a model from scratch while using a unified one. All of them have not yet made full use of the already well-trained text detectors and easily obtainable detection datasets. In this paper, we present Text Grouping Adapter (TGA), a module that can enable the utilization of various pretrained text detectors to learn layout analysis, allowing us to adopt a well-trained text detector right off the shelf or just fine-tune it efficiently. Designed to be compatible with various text detector architectures, TGA takes detected text regions and image features as universal inputs to as-semble text instance features. To capture broader contextual information for layout analysis, we propose to predict text group masks from text instance features by one-to-many assignment. Our comprehensive experiments demonstrate that, even with frozen pretrained models, incorporating our TGA into various pretrained text detectors and text spotters can achieve superior layout analysis performance, simultaneously inheriting generalized text detection ability from pretraining. In the case of full parameter fine-tuning, we can further improve layout analysis performance.

Recent advancements in video modeling extensively rely on optical flow to represent the relationships across frames, but this approach often lacks efficiency and fails to model the probability of the intrinsic motion of objects. In addition, conventional encoder-decoder frameworks in video processing focus on modeling the correlation in the encoder, leading to limited generative capabilities and redundant intermediate representations. To address these challenges, this paper proposes a novel Implicit Motion Function (IMF) method. Our approach utilizes a low-dimensional latent token as the implicit representation, along with the use of cross-attention, to implicitly model the correlation between frames. This enables the implicit modeling of temporal correlations and understanding of object motions. Our method not only improves sparsity and efficiency in representation but also explores the generative capabilities of the decoder by integrating correlation modeling within it. The IMF framework facilitates video editing and other generative tasks by allowing the direct manipulation of latent tokens. We validate the effectiveness of IMF through extensive experiments on multiple video tasks, demonstrating superior performance in terms of reconstructed video quality, compression efficiency and generation ability.

Most existing image compression approaches perform transform coding in the pixel space to reduce its spatial re-dundancy. However, they encounter difficulties in achieving both high-realism and high-fidelity at low bitrate, as the pixel-space distortion may not align with human perception. To address this issue, we introduce a Generative Latent Coding (GLC) architecture, which performs transform coding in the latent space of a generative vector-quantized variational auto-encoder (VQ- VAE), instead of in the pixel space. The generative latent space is characterized by greater sparsity, richer semantic and better alignment with human perception, rendering it advantageous for achieving high-realism and high-fidelity compression. Additionally, we introduce a categorical hyper module to reduce the bit cost of hyper-information, and a code-prediction-based su-pervision to enhance the semantic consistency. Experiments demonstrate that our GLC maintains high visual quality with less than 0.04 bpp on natural images and less than 0.01 bpp on facial images. On the CLIC2020 test set, we achieve the same FID as MS-ILLM with 45% fewer bits. Furthermore, the powerful generative latent space enables various applications built on our GLC pipeline, such as image restoration and style transfer.

Recent methods for label-free 3D semantic segmentation aim to assist 3D model training by leveraging the open-world recognition ability of pre-trained vision language models. However, these methods usually suffer from in-consistent and noisy pseudo-labels provided by the vision language models. To address this issue, we present a hierarchical intra-modal correlation learning framework that captures visual and geometric correlations in 3D scenes at three levels: intra-set, intra-scene, and inter-scene, to help learn more compact 3D representations. We refine pseudo-labels using intra-set correlations within each geometric consistency set and align features of visually and geometrically similar points using intra-scene and inter-scene correlation learning. We also introduce a feedback mechanism to distill the correlation learning capability into the 3D model. Experiments on both indoor and outdoor datasets show the superiority of our method. We achieve a state-of-the-art 36.6% mIoU on the ScanNet dataset, and a 23.0% mIoU on the nuScenes dataset, with improvements of 7.8% mIoU and 2.2% mIoU compared with previous SOTA. We also provide theoretical analysis and qualitative visualization results to discuss the mechanism and conduct thorough ablation studies to support the effectiveness of our framework.

The emerging conditional coding-based neural video codec (NVC) shows superiority over commonly-used resid-ual coding-based codec and the latest NVC already claims to outperform the best traditional codec. However, there still exist critical problems blocking the practicality of NVC. In this paper, we propose a powerful conditional coding- based NVC that solves two critical problems via feature modulation. The first is how to support a wide quality range in a single model. Previous NVC with this capability only supports about 3.8 dB PSNR range on average. To tackle this limitation, we modulate the latent feature of the cur-rent frame via the learnable quantization scaler. During the training, we specially design the uniform quantization pa-rameter sampling mechanism to improve the harmonization of encoding and quantization. This results in a better learning of the quantization scaler and helps our NVC support about 11.4 dB PSNR range. The second is how to make NVC still work under a long prediction chain. We expose that the previous SOTA NVC has an obvious quality degra-dation problem when using a large intra-period setting. To this end, we propose modulating the temporal feature with a periodically refreshing mechanism to boost the quality. Notably, under single intra-frame setting, our codec can achieve 29.7% bitrate saving over previous SOTA NVC with 16% MACs reduction. Our codec serves as a notable land-mark in the journey of NVC evolution. The codes are at https://github.com/microsoft/DCVC.

Recently, integrating video foundation models and large language models to build a video understanding system can overcome the limitations of specific pre-defined vision tasks. Yet, existing systems can only handle videos with very few frames. For long videos, the computation complexity, memory cost, and long-term temporal connection impose additional challenges. Taking advantage of the Atkinson-Shiffrin memory model, with tokens in Transformers being employed as the carriers of memory in combination with our specially designed memory mechanism, we propose the MovieChat to overcome these challenges. MovieChat achieves state-of-the-art performance in long video understanding, along with the released MovieChat-1K benchmark with 1K long video and 14K manual annotations for validation of the effectiveness of our method. The code, models and data can be found in https://reself.github.io/MovieChat.

Existing deep learning based speech enhancement mainly employ a data-driven approach, which leverage large amounts of data with a variety of noise types to achieve noise removal from noisy signal. However, the high dependence on the data limits its generalization on the unseen complex noises in real-life environment. In this paper, we focus on the low-latency scenario and regard speech enhancement as a speech generation problem conditioned on the noisy signal, where we generate clean speech instead of identifying and removing noises. Specifically, we propose a conditional generative framework for speech enhancement, which models clean speech by acoustic codes of a neural speech codec and generates the speech codes conditioned on past noisy frames in an auto-regressive way. Moreover, we propose an explicitalignment approach to align noisy frames with the generated speech tokens to improve the robustness and scalability to different input lengths. Different from other methods that leverage multiple stages to generate speech codes, we leverage a single-stage speech generation approach based on the TF-Codec neural codec to achieve high speech quality with low latency. Extensive results on both synthetic and real-recorded test set show its superiority over data-driven approaches in terms of noise robustness and temporal speech coherence.

Most Neural Radiance Fields (NeRFs) exhibit limited generalization capabilities,which restrict their applicability in representing multiple scenes using a single model. To address this problem, existing generalizable NeRF methods simply condition the model on image features. These methods still struggle to learn precise global representations over diverse scenes since they lack an effective mechanism for interacting among different points and views. In this work, we unveil that 3D implicit representation learning can be significantly improved by mask-based modeling. Specifically, we propose **m**asked **r**ay and **v**iew **m**odeling for generalizable **NeRF** (**MRVM-NeRF**), which is a self-supervised pretraining target to predict complete scene representations from partially masked features along each ray. With this pretraining target, MRVM-NeRF enables better use of correlations across different rays and views as the geometry priors, which thereby strengthens the capability of capturing intricate details within the scenes and boosts the generalization capability across different scenes. Extensive experiments demonstrate the effectiveness of our proposed MRVM-NeRF on both synthetic and real-world datasets, qualitatively and quantitatively. Besides, we also conduct experiments to show the compatibility of our proposed method with various backbones and its superiority under few-shot cases.

In-context learning, i.e., learning from context examples, is an impressive ability of Transformer. Training Transformers to possess this in-context learning skill is computationally intensive due to the occurrence of *learning plateaus*, which are periods within the training process where there is minimal or no enhancement in the model's in-context learning capability. To study the mechanism behind the learning plateaus, we conceptually separate a component within the model's internal representation that is exclusively affected by the model's weights. We call this the “weights component”, and the remainder is identified as the “context component”. By conducting meticulous and controlled experiments on synthetic tasks, we note that the persistence of learning plateaus correlates with compromised functionality of the weights component. Recognizing the impaired performance of the weights component as a fundamental behavior that drives learning plateaus, we have developed three strategies to expedite the learning of Transformers. The effectiveness of these strategies is further confirmed in natural language processing tasks. In conclusion, our research demonstrates the feasibility of cultivating a powerful in-context learning ability within AI systems in an eco-friendly manner.

Low resource of parallel data is the key challenge of accent conversion(AC) problem in which both the pronunciation units and prosody pattern need to be converted. We propose a two-stage generative framework "convert-and-speak" in which the conversion is only operated on the semantic token level and the speech is synthesized conditioned on the converted semantic token with a speech generative model in target accent domain. The decoupling design enables the "speaking" module to use massive amount of target accent speech and relieves the parallel data required for the "conversion" module. Conversion with the bridge of semantic token also relieves the requirement for the data with text transcriptions and unlocks the usage of language pre-training technology to further efficiently reduce the need of parallel accent speech data. To reduce the complexity and latency of "speaking", a single-stage AR generative model is designed to achieve good quality as well as lower computation cost. Experiments on Indian-English to general American-English conversion show that the proposed framework achieves state-of-the-art performance in accent similarity, speech quality, and speaker maintenance with only 15 minutes of weakly parallel data which is not constrained to the same speaker. Extensive experimentation with diverse accent types suggests that this framework possesses a high degree of adaptability, making it readily scalable to accommodate other accents with low-resource data. Audio samples are available at https://www.microsoft.com/en-us/research/project/convert-and-speak-zero-shot-accent-conversion-with-minimumsupervision/.

Video-Language Models (VLMs), powered by the advancements in Large Language Models (LLMs), are charting new frontiers in video understanding. A pivotal challenge is the development of an effective method to encapsulate video content into a set of representative tokens to align with LLMs. In this work, we introduce Slot-VLM, a new framework designed to generate semantically decomposed video tokens, in terms of object-wise and event-wise visual representations, to facilitate LLM inference. Particularly, we design an Object-Event Slots module, i.e., OE-Slots, that adaptively aggregates the dense video tokens from the vision encoder to a set of representative slots. In order to take into account both the spatial object details and the varied temporal dynamics, we build OE-Slots with two branches: the Object-Slots branch and the Event-Slots branch. The Object-Slots branch focuses on extracting object-centric slots from features of high spatial resolution but low frame sample rate, emphasizing detailed object information. The Event-Slots branch is engineered to learn event-centric slots from high temporal sample rate but low spatial resolution features. These complementary slots are combined to form the vision context, serving as the input to the LLM for effective video reasoning. Our experimental results demonstrate the effectiveness of our Slot-VLM, which achieves the state-of-the-art performance on video question-answering.

Disentangled representation learning strives to extract the intrinsic factors within the observed data. Factoring these representations in an unsupervised manner is notably challenging and usually requires tailored loss functions or specific structural designs. In this paper, we introduce a new perspective and framework, demonstrating that diffusion models with cross-attention itself can serve as a powerful inductive bias to facilitate the learning of disentangled representations. We propose to encode an image into a set of concept tokens and treat them as the condition of the latent diffusion model for image reconstruction, where cross attention over the concept tokens is used to bridge the encoder and the U-Net of the diffusion model. We analyze that the diffusion process inherently possesses the time-varying information bottlenecks. Such information bottlenecks and cross attention act as strong inductive biases for promoting disentanglement. Without any regularization term in the loss function, this framework achieves superior disentanglement performance on the benchmark datasets, surpassing all previous methods with intricate designs. We have conducted comprehensive ablation studies and visualization analyses, shedding a light on the functioning of this model. We anticipate that our findings will inspire more investigation on exploring diffusion model for disentangled representation learning towards more sophisticated data analysis and understanding.

Tracking visual objects from a single initial exemplar in the testing phase has been broadly cast as a one-/few-shot problem, i.e., one-shot learning for initial adaptation and few-shot learning for online adaptation. The recent few-shot online adaptation methods incorporate the prior knowledge from large amounts of annotated training data via complex meta-learning optimization in the offline phase. This helps the online deep trackers to achieve fast adaptation and reduce overfitting risk in tracking. In this paper, we propose a simple yet effective recursive least-squares estimator-aided online learning approach for few-shot online adaptation without requiring offline training. It allows an in-built memory retention mechanism for the model to remember the knowledge about the object seen before, and thus the seen data can be safely removed from training. This also bears certain similarities to the emerging continual learning field in preventing catastrophic forgetting. This mechanism enables us to unveil the power of modern online deep trackers without incurring too much extra computational cost. We evaluate our approach based on two networks in the online learning families for tracking, i.e., multi-layer perceptrons in RT-MDNet and convolutional neural networks in DiMP. The consistent improvements on several challenging tracking benchmarks demonstrate its effectiveness and efficiency.

Deep learning-based video compression is a challenging task, and many previous state-of-the-art learning-based video codecs use optical flows to exploit the temporal correlation between successive frames and then compress the residual error. Although these two-stage models are end-to-end optimized, the epistemic uncertainty in the motion estimation and the aleatoric uncertainty from the quantization operation lead to errors in the intermediate representations and introduce artifacts in the reconstructed frames. This inherent flaw limits the potential for higher bit rate savings. To address this issue, we propose an uncertainty-aware video compression model that can effectively capture the predictive uncertainty with deep ensembles. Additionally, we introduce an ensemble-aware loss to encourage the diversity among ensemble members and investigate the benefits of incorporating adversarial training in the video compression task. Experimental results on 1080p sequences show that our model can effectively save bits by more than 20% compared to DVC Pro.

In recent years, considerable progress has been witnessed in the person re-identification (Re-ID). However, in a more realistic long-term scenario, the appearance shift arising from the clothes-changing inevitably deteriorates the conventional methods that heavily depend on the clothing color. Although the current clothes-changing person Re-ID methods introduce external human knowledge (i.e, contour, mask) and sophisticated feature decoupling strategy to alleviate the clothing shift, they still face the risk of overfitting to clothing due to the limited clothing diversity of training set. To more efficiently and effectively promote the clothes-irrelevant feature learning, we present a novel joint Identity-aware Mixstyle and Graph-enhanced Prototype method for clothes-changing person Re-ID. Specifically, by treating the cloth-changing as fine-grained domain/style shift, the identity-aware mixstyle (IMS) is proposed from the perspective of domain generalization, which mixes the instance-level feature statistics of samples within each identity to synthesize novel and diverse clothing styles, while retaining the correspondence between synthesized samples and latent label space. By incorporating the IMS module, the more diverse styles can be exploited to train a clothing-shift robust model. To further reduce the feature discrepancy caused by clothing variations, the graph-enhanced prototype constraint (GEP) module is proposed to explore the graph similarity structure of style-augmented samples across memory bank to build informative and robust prototypes, which serve as powerful exemplars for better clothing-irrelevant metric learning. The two modules are integrated into a joint learning framework and benefit each other. The extensive experiments conducted on clothes-changing person Re-ID datasets validate the superiority and effectiveness of our method. In addition, our method also shows good universality and corruption robustness on other Re-ID tasks.

Video frame synthesis, which consists of interpolation and extrapolation , is an essential video processing technique that can be applied to various scenarios. However, most existing methods cannot handle small objects or large motion well, especially in high-resolution videos such as 4K videos. To eliminate such limitations, we introduce a neighbor correspondence matching (NCM) algorithm for flow-based frame synthesis. Since the current frame is not available in video frame synthesis, NCM is performed in a current-frame-agnostic fashion to establish multi-scale correspondences in the spatial-temporal neighborhoods of each pixel. Based on the powerful motion representation capability of NCM, we propose a heterogeneous coarse-to-fine scheme for intermediate flow estimation. The coarse-scale and fine-scale modules are trained progressively, making NCM computationally efficient and robust to large motions. We further explore the mechanism of NCM and find that neighbor correspondence is powerful, since it provides multiple-hypotheses motion information for synthesis. Based on this analysis, we introduce a multiple-hypotheses estimation process for video frame extrapolation, resulting in a more robust framework, NCM-MH. Experimental results show that NCM and NCM-MH achieve 31.63 and 28.08 dB for interpolation and extrapolation on the most challenging X4K1000FPS benchmark, outperforming all the other state-of-the-art methods that use two reference frames as input.

Recently, machine learning-based image compression has attracted increasing interests and is approaching the state-of-the-art compression ratio. But unlike traditional codec, it lacks a universal optimization method to seek efficient representation for different images. In this paper, we develop a plug-and-play optimization framework for seeking higher compression ratio, which can be flexibly applied to existing and potential future compression networks. To make the latent representation more efficient, we propose a novel latent optimization algorithm to adaptively remove the redundancy for each image. Additionally, inspired by the potential of side information for traditional codecs, we introduce side information into our framework, and integrate side information optimization with latent optimization to further enhance the compression ratio. In particular, with the joint side information and latent optimization, we can achieve fine rate control using only single model instead of training different models for different rate-distortion trade-offs, which significantly reduces the training and storage cost to support multiple bit rates. Experimental results demonstrate that our proposed framework can remarkably boost the machine learning-based compression ratio, achieving more than 10% additional bit rate saving on three different representative network structures. With the proposed optimization framework, we can achieve 7.6% bit rate saving against the latest traditional coding standard VVC on Kodak dataset, yielding the state-of-the-art compression ratio.

The three existing dominant network families, i.e., CNNs, Transformers and MLPs, differ from each other mainly in the ways of fusing spatial contextual information, leaving designing more effective token-mixing mechanisms at the core of backbone architecture development. In this work, we propose an innovative token-mixer, dubbed Active Token Mixer (ATM), to actively incorporate contextual information from other tokens in the global scope into the given query token. This fundamental operator actively predicts where to capture useful contexts and learns how to fuse the captured contexts with the query token at channel level. In this way, the spatial range of token-mixing can be expanded to a global scope with limited computational complexity, where the way of token-mixing is reformed. We take ATMs as the primary operators and assemble them into a cascade architecture, dubbed ATMNet. Extensive experiments demonstrate that ATMNet is generally applicable and comprehensively surpasses different families of SOTA vision backbones by a clear margin on a broad range of vision tasks, including visual recognition and dense prediction tasks. Code is available at https://github.com/microsoft/ActiveMLP.

The Multiplane Image (MPI), containing a set of fronto-parallel$RGB_{\alpha}$layers, is an effective and efficient representation for view synthesis from sparse inputs. Yet, its fixed structure limits the performance, especially for surfaces imaged at oblique angles. We introduce the Structural MPI (S-MPI), where the plane structure approximates 3D scenes concisely. Conveying$RGB_{\alpha}$contexts with geometrically-faithful structures, the S-MPI directly bridges view synthe-sis and 3D reconstruction. It can not only overcome the critical limitations of MPI, i.e., discretization artifacts from sloped surfaces and abuse of redundant layers, and can also acquire planar 3D reconstruction. Despite the intu-ition and demand of applying S-MPI, great challenges are introduced,$e.g$., high-fidelity approximation for both$RGB_{\alpha}$layers and plane poses, multi-view consistency, non-planar regions modeling, and efficient rendering with intersected planes. Accordingly, we propose a transformer-based network based on a segmentation model [4]. It predicts compact and expressive S-MPI layers with their corresponding masks, poses, and$RGB_{\alpha}$contexts. Non-planar regions are inclusively handled as a special case in our unified frame-work. Multi-view consistency is ensured by sharing global proxy embeddings, which encode plane-level features cov-ering the complete 3D scenes with aligned coordinates. In-tensive experiments show that our method outperforms both previous state-of-the-art MPI-based view synthesis methods and planar reconstruction methods.

Talking head generation is to generate video based on a given source identity and target motion. However, current methods face several challenges that limit the quality and controllability of the generated videos. First, the generated face often has unexpected deformation and severe distortions. Second, the driving image does not explicitly disentangle movement-relevant information, such as poses and expressions, which restricts the manipulation of different attributes during generation. Third, the generated videos tend to have flickering artifacts due to the inconsistency of the extracted landmarks between adjacent frames. In this paper, we propose a novel model that produces high-fidelity talking head videos with free control over head pose and expression. Our method leverages both self-supervised learned landmarks and 3D face model-based landmarks to model the motion. We also introduce a novel motion-aware multi-scale feature alignment module to effectively transfer the motion without face distortion. Furthermore, we enhance the smoothness of the synthesized talking head videos with a feature context adaptation and propagation module. We evaluate our model on challenging datasets and demonstrate its state-of-the-art performance. More information is available at https://yuegao.me/PECHead.

Layout generation aims to synthesize realistic graphic scenes consisting of elements with different attributes in-cluding category, size, position, and between-element relation. It is a crucial task for reducing the burden on heavyduty graphic design works for formatted scenes, e.g., publications, documents, and user interfaces (UIs). Diverse application scenarios impose a big challenge in unifying various layout generation subtasks, including conditional and unconditional generation. In this paper, we propose a Layout Diffusion Generative Model (LDGM) to achieve such unification with a single decoupled diffusion model. LDGM views a layout of arbitrary missing or coarse element attributes as an intermediate diffusion status from a completed layout. Since different attributes have their individual semantics and characteristics, we propose to decouple the diffusion processes for them to improve the diversity of training samples and learn the reverse process jointly to exploit global-scope contexts for facilitating generation. As a result, our LDGM can generate layouts either from scratch or conditional on arbitrary available attributes. Extensive qualitative and quantitative experiments demonstrate our proposed LDGM outperforms existing layout generation models in both functionality and performance.

For any video codecs, the coding efficiency highly relies on whether the current signal to be encoded can find the relevant contexts from the previous reconstructed signals. Traditional codec has verified more contexts bring substantial coding gain, but in a time-consuming manner. However, for the emerging neural video codec (NVC), its contexts are still limited, leading to low compression ratio. To boost NVC, this paper proposes increasing the context diversity in both temporal and spatial dimensions. First, we guide the model to learn hierarchical quality patterns across frames, which enriches long-term and yet highquality temporal contexts. Furthermore, to tap the potential of optical flow-based coding framework, we introduce a group-based offset diversity where the cross-group interaction is proposed for better context mining. In addition, this paper also adopts a quadtree-based partition to increase spatial context diversity when encoding the latent representation in parallel. Experiments show that our codec obtains 23.5% bitrate saving over previous SOTA NVC. Better yet, our codec has surpassed the under-developing next generation traditional codec/ECM in both RGB and YUV420 colorspaces, in terms of PSNR. The codes are at https://github.com/microsoft/DCVC.

Improving the generalization ability of Deep Neural Networks (DNNs) is critical for their practical uses, which has been a longstanding challenge. Some theoretical studies have uncovered that DNNs have preferences for some frequency components in the learning process and indicated that this may affect the robustness of learned features. In this paper, we propose Deep Frequency Filtering (DFF)for learning domain-generalizable features, which is the first endeavour to explicitly modulate the frequency components of different transfer difficulties across domains in the latent space during training. To achieve this, we perform Fast Fourier Transform (FFT) for the feature maps at different layers, then adopt a light-weight module to learn attention masks from the frequency representations after FFT to enhance transferable components while suppressing the components not conducive to generalization. Further, we empirically compare the effectiveness of adopting different types of attention designs for implementing DFF. Extensive experiments demonstrate the effectiveness of our proposed DFF and show that applying our DFF on a plain baseline out-performs the state-of-the-art methods on different domain generalization tasks, including close-set classification and open-set retrieval.

In most existing neural video codecs, the information flow therein is uni-directional, where only motion coding provides motion vectors for frame coding. In this paper, we argue that, through information interactions, the synergy between motion coding and frame coding can be achieved. We effectively introduce bi-directional information interactions between motion coding and frame coding via our Motion Information Propagation. When generating the temporal contexts for frame coding, the high-dimension motion feature from the motion decoder serves as motion guidance to mitigate the alignment errors. Meanwhile, besides assisting frame coding at the current time step, the feature from context generation will be propagated as motion condition when coding the subsequent motion latent. Through the cycle of such interactions, feature propagation on motion coding is built, strengthening the capacity of exploiting long-range temporal correlation. In addition, we propose hybrid context generation to exploit the multiscale context features and provide better motion condition. Experiments show that our method can achieve 12.9% bit rate saving over the previous SOTA neural video codec.

Existing image captioning methods are under the assumption that the training and testing data are from the same domain or that the data from the target domain (i.e., the domain that testing data lie in) are accessible. However, this assumption is invalid in real-world applications where the data from the target domain is inaccessible. In this paper, we introduce a new setting called Domain Generalization for Image Captioning (DGIC), where the data from the target domain is unseen in the learning process. We first construct a benchmark dataset for DGIC, which helps us to investigate models' domain generalization (DG) ability on unseen domains. With the support of the new benchmark, we further propose a new framework called language-guided semantic metric learning (LSML) for the DGIC setting. Experiments on multiple datasets demonstrate the challenge of the task and the effectiveness of our newly proposed benchmark and LSML framework.

Existing Siamese tracking methods, which are built on pair-wise matching between two single frames, heavily rely on additional sophisticated mechanism to exploit temporal information among successive video frames, hindering them from efficiency and industrial deployments. In this work, we resort to sequence-level target matching that can encode temporal contexts into the spatial features through a neat feedforward video model. Specifically, we adapt the standard video transformer architecture to visual tracking by enabling spatiotemporal feature learning directly from frame-level patch sequences. To better adapt to the tracking task, we carefully blend the spatiotemporal information in the video clips through sequential multi-branch triplet blocks, which formulates a video transformer backbone. Our experimental study compares different model variants, such as tokenization strategies, hierarchical structures, and video attention schemes. Then, we propose a disentangled dual-template mechanism that decouples static and dynamic appearance clues over time, and reduces temporal redundancy in video frames. Extensive experiments show that our method, named as Video Track, achieves state-of-the-art results while running in real-time.

Previous works on video object segmentation (VOS) are trained on densely annotated videos. Nevertheless, acquiring annotations in pixel level is expensive and time-consuming. In this work, we demonstrate the feasibility of training a satisfactory VOS model on sparsely annotated videos—we merely require two labeled frames per training video while the performance is sustained. We term this novel training paradigm as two-shot video object segmentation, or two-shot VOS for short. The underlying idea is to generate pseudo labels for unlabeled frames during training and to optimize the model on the combination of labeled and pseudo-labeled data. Our approach is extremely simple and can be applied to a majority of existing frameworks. We first pre-train a VOS model on sparsely annotated videos in a semi-supervised manner, with the first frame always being a labeled one. Then, we adopt the pretrained VOS model to generate pseudo labels for all unlabeled frames, which are subsequently stored in a pseudo-label bank. Finally, we retrain a VOS model on both labeled and pseudo-labeled data without any restrictions on the first frame. For the first time, we present a general way to train VOS models on two-shot VOS datasets. By using 7.3% and 2.9% labeled data of YouTube-VOS and DAVIS benchmarks, our approach achieves comparable results in contrast to the counterparts trained on fully labeled set. Code and models are available at https://github.com/yk-pku/Two-shot-Video-Object-Segmentation.

Recently end-to-end neural audio/speech coding has shown its great potential to outperform traditional signal analysis based audio codecs. This is mostly achieved by following the VQ-VAE paradigm where blind features are learned, vector-quantized and coded. In this paper, instead of blind end-to-end learning, we propose to learn disentangled features for real-time neural speech coding. Specifically, more global-like speaker identity and local content features are learned with disentanglement to represent speech. Such a compact feature decomposition not only achieves better coding efficiency by exploiting bit allocation among different features but also provides the flexibility to do audio editing in embedding space, such as voice conversion in real-time communications. Both subjective and objective results demonstrate its coding efficiency and we find that the learned disentangled features show comparable performance on any-to-any voice conversion with modern self-supervised speech representation learning models with far less parameters and low latency, showing the potential of our neural coding framework.

For the task of speech separation, previous study usually treats multi-channel and single-channel scenarios as two research tracks with specialized solutions developed respectively. Instead, we propose a simple and unified architecture - DasFormer (Deep alternating spectrogram transFormer) to handle both of them in the challenging reverberant environments. Unlike frame-wise sequence modeling, each TF-bin in the spectrogram is assigned with an embedding encoding spectral and spatial information. With such input, DasFormer is then formed by multiple repetition of simple blocks each of which integrates 1) two multi-head self-attention (MHSA) modules alternately processing within each frequency bin & temporal frame of the spectrogram 2) MBConv before each MHSA for modeling local features on the spectrogram. Experiments show that DasFormer has a powerful ability to model the time-frequency representation, whose performance far exceeds the current SOTA models in multi-channel speech separation, and also achieves single-channel SOTA in the more challenging yet realistic reverberation scenario.

Existing deep learning based speech enhancement (SE) methods either use blind end-to-end training or explicitly incorporate speaker embedding or phonetic information into the SE network to enhance speech quality. In this paper, we perceive speech and noises as different types of sound events and propose an event-based query method for SE. Specifically, speech embeddings that can discriminate speech from noises are first pre-trained with the sound event detection (SED) task. The embeddings are then clustered into fixed golden speech queries, i.e., general but representative speech embeddings, on a diverse clean speech dataset to assist the SE network. The golden speech queries can be obtained offline and generalizable to different SE datasets and networks. Therefore, little extra complexity is introduced and no enrollment is needed for each speaker. Experimental results show that the proposed method yields significant gains compared with baselines and the golden queries are well generalized to different datasets.

Packet loss concealment (PLC) is challenging in concealing missing contents both plausibly and naturally when there are only limited available context to use. Recently deep-learning based PLC algorithms have demonstrated their superiority over traditional counterparts; but their concealment ability is still mostly limited to a maximum of 120ms loss. Even with strong GAN-based generative models, it is still very challenging to predict long burst losses that could happen within/in-between phonemes. In this paper, we propose to use contrastive learning to learn a loss-robust semantic representation for PLC. A hybrid neural PLC architecture combining the semantic prediction and GAN-based generative model is designed to verify its effectiveness. Results on the blind test set of Interspeech2022 PLC Challenge show its superiority over commonly used UNet-style framework and the one without contrastive learning, especially for the longer burst loss at (120, 220]ms.

Transformer is popular in recent 3D human pose estimation, which utilizes long-term modeling to lift 2D keypoints into the 3D space. However, current transformer-based methods do not fully exploit the prior knowledge of the human skeleton provided by the kinematic structure. In this paper, we propose a novel transformer-based model EvoPose to introduce the human body prior knowledge for 3D human pose estimation effectively. Specifically, a Structural Priors Representation (SPR) module represents human priors as structural features carrying rich body patterns, e.g. joint relationships. The structural features are interacted with 2D pose sequences and help the model to achieve more informative spatiotemporal features. Moreover, a Recursive Refinement (RR) module is applied to refine the 3D pose outputs by utilizing estimated results and further injects human priors simultaneously. Extensive experiments demonstrate the effectiveness of EvoPose which achieves a new state of the art on two most popular benchmarks, Human3.6M and MPI-INF-3DHP.

For real-time speech enhancement (SE) including noise suppression, dereverberation and acoustic echo cancellation, the time-variance of the audio signals becomes a severe challenge. The causality and memory usage limit that only the historical information can be used for the system to capture the time-variant characteristics. We propose to adaptively change the receptive field according to the input signal in deep neural network based SE model. Specifically, in an encoder-decoder framework, a dynamic attention span mechanism is introduced to all the attention modules for controlling the size of historical content used for processing the current frame. Experimental results verify that this dynamic mechanism can better track time-variant factors and capture speech-related characteristics, benefiting to both interference removing and speech quality retaining.

Recent vision transformers, large-kernel CNNs and MLPs have attained remarkable successes in broad vision tasks thanks to their effective information fusion in the global scope. However, their efficient deployments, especially on mobile devices, still suffer from noteworthy challenges due to the heavy computational costs of self-attention mechanisms, large kernels, or fully connected layers. In this work, we apply conventional convolution theorem to deep learning for addressing this and reveal that adaptive frequency filters can serve as efficient global token mixers. With this insight, we propose Adaptive Frequency Filtering (AFF) token mixer. This neural operator transfers a latent representation to the frequency domain via a Fourier transform and performs semantic-adaptive frequency filtering via an elementwise multiplication, which mathematically equals to a token mixing operation in the original latent space with a dynamic convolution kernel as large as the spatial resolution of this latent representation. We take AFF token mixers as primary neural operators to build a lightweight neural network, dubbed AFFNet. Extensive experiments demonstrate the effectiveness of our proposed AFF token mixer and show that AFFNet achieve superior accuracy and efficiency trade-offs compared to other lightweight network designs on broad visual tasks, including visual recognition and dense prediction tasks. Code is available at https://github.com/microsoft/TokenMixers.

Diffusion-based methods can generate realistic images and videos, but they struggle to edit existing objects in a video while preserving their appearance over time. This prevents diffusion models from being applied to natural video editing in practical scenarios. In this paper, we tackle this problem by introducing temporal dependency to existing text-driven diffusion models, which allows them to generate consistent appearance for the edited objects. Specifically, we develop a novel inter-frame propagation mechanism for diffusion video editing, which leverages the concept of layered representations to propagate the appearance information from one frame to the next. We then build up a text-driven video editing framework based on this mechanism, namely StableVideo, which can achieve consistency-aware video editing. Extensive experiments demonstrate the strong editing capability of our approach. Compared with state-of-the-art video editing methods, our approach shows superior qualitative and quantitative results. Our code is available at this https URL.

Referring Video Object Segmentation (R-VOS) is a challenging task that aims to segment an object in a video based on a linguistic expression. Most existing R-VOS methods have a critical assumption: the object referred to must appear in the video. This assumption, which we refer to as "semantic consensus", is often violated in real-world scenarios, where the expression may be queried against false videos. In this work, we highlight the need for a robust R-VOS model that can handle semantic mismatches. Accordingly, we propose an extended task called Robust R-VOS (R2-VOS), which accepts unpaired video-text inputs. We tackle this problem by jointly modeling the primary R-VOS problem and its dual (text reconstruction). A structural text-to-text cycle constraint is introduced to discriminate semantic consensus between video-text pairs and impose it in positive pairs, thereby achieving multi-modal alignment from both positive and negative pairs. Our structural constraint effectively addresses the challenge posed by linguistic diversity, overcoming the limitations of previous methods that relied on the point-wise constraint. A new evaluation dataset, R2-Youtube-VOS is constructed to measure the model robustness. Our model achieves state-of-the-art performance on R-VOS benchmarks, Ref-DAVIS17 and Ref-Youtube-VOS, and also our R2-Youtube-VOS dataset.

Recent work on Neural Radiance Fields (NeRF) has demonstrated significant advances in high-quality view synthesis. A major limitation of NeRF is its low rendering efficiency due to the need for multiple network forwardings to render a single pixel. Existing methods to improve NeRF either reduce the number of required samples or optimize the implementation to accelerate the network forwarding. Despite these efforts, the problem of multiple sampling persists due to the intrinsic representation of radiance fields. In contrast, Neural Light Fields (NeLF) reduce the computation cost of NeRF by querying only one single network forwarding per pixel. To achieve a close visual quality to NeRF, existing NeLF methods require significantly larger network capacities which limits their rendering efficiency in practice. In this work, we propose a new representation called Neural Radiance Distribution Field (NeRDF) that targets efficient view synthesis in real-time. Specifically, we use a small network similar to NeRF while preserving the rendering speed with a single network forwarding per pixel as in NeLF. The key is to model the radiance distribution along each ray with frequency basis and predict frequency weights using the network. Pixel values are then computed via volume rendering on radiance distributions. Experiments show that our proposed method offers a better trade-off among speed, quality, and network size than existing methods: we achieve a ∼254× speed-up over NeRF with similar network size, with only a marginal performance decline. Our project page is at yushuang-wu.github.io/NeRDF.

Noise suppression (NS) models have been widely applied to enhance speech quality.Recently, Deep Learning-Based NS, which we denote as Deep Noise Suppression (DNS), became the mainstream NS method due to its excelling performance over traditional ones.However, DNS models face 2 major challenges for supporting the real-world applications.First, highperforming DNS models are usually large in size, causing deployment difficulties.Second, DNS models require extensive training data, including noisy audios as inputs and clean audios as labels.It is often difficult to obtain clean labels for training DNS models.We propose the use of knowledge distillation (KD) to resolve both challenges.Our study serves 2 main purposes.To begin with, we are among the first to comprehensively investigate mainstream KD techniques on DNS models to resolve the two challenges.Furthermore, we propose a novel Attention-Based-Compression KD method that outperforms all investigated mainstream KD frameworks on DNS task.

We propose the first framework for accelerating video recovery, which aims to efficiently recover high-quality videos from degraded inputs affected by various deteriorative factors. Although current video recovery methods have achieved excellent performance, their significant computational overhead limits their widespread application. To address this, we present a pioneering study on explicitly disentangling temporal and spatial redundant computation by decomposing the input frame into propagation and restoration regions, thereby achieving significant computational reduction. Specifically, we leverage contrastive learning to learn degradation-invariant features, which overcomes the disturbance of deteriorative factors and enables accurate disentanglement. For the propagation region, we introduce a split-fusion block to address inter-frame variations, efficiently generating high-quality output at a low cost and significantly reducing temporal redundant computation. For the restoration region, we propose an efficient adaptive halting mechanism that requires few extra parameters and can adaptively halt the patch processing, considerably reducing spatial redundant computation. Furthermore, we design patch-adaptive prior regularization to boost efficiency and performance. Our proposed method achieves outstanding results on various video recovery tasks, such as video denoising, video deraining, video dehazing, and video super-resolution, with a 50% ~ 60% reduction in GMAC over the state-of-the-art video recovery methods while maintaining comparable performance.

This paper explores the connection between learning trajectories of Deep Neural Networks (DNNs) and their generalization capabilities when optimized using (stochastic) gradient descent algorithms. Instead of concentrating solely on the generalization error of the DNN post-training, we present a novel perspective for analyzing generalization error by investigating the contribution of each update step to the change in generalization error. This perspective enable a more direct comprehension of how the learning trajectory influences generalization error. Building upon this analysis, we propose a new generalization bound that incorporates more extensive trajectory information. Our proposed generalization bound depends on the complexity of learning trajectory and the ratio between the bias and diversity of training set. Experimental observations reveal that our method effectively captures the generalization error throughout the training process. Furthermore, our approach can also track changes in generalization error when adjustments are made to learning rates and label noise levels. These results demonstrate that learning trajectory information is a valuable indicator of a model's generalization capabilities.

Targeting to understand the underlying explainable factors behind observations and modeling the conditional generation process on these factors, we connect disentangled representation learning to diffusion probabilistic models (DPMs) to take advantage of the remarkable modeling ability of DPMs. We propose a new task, disentanglement of (DPMs): given a pre-trained DPM, without any annotations of the factors, the task is to automatically discover the inherent factors behind the observations and disentangle the gradient fields of DPM into sub-gradient fields, each conditioned on the representation of each discovered factor. With disentangled DPMs, those inherent factors can be automatically discovered, explicitly represented and clearly injected into the diffusion process via the sub-gradient fields. To tackle this task, we devise an unsupervised approach, named DisDiff, and for the first time achieving disentangled representation learning in the framework of DPMs. Extensive experiments on synthetic and real-world datasets demonstrate the effectiveness of DisDiff.

Neural audio/speech coding has recently demonstrated its capability to deliver high quality at much lower bitrates than traditional methods. However, existing neural audio/speech codecs employ either acoustic features or learned blind features with a convolutional neural network for encoding, by which there are still temporal redundancies within encoded features. This article introduces latent-domain predictive coding into the VQ-VAE framework to fully remove such redundancies and proposes the TF-Codec for low-latency neural speech coding in an end-to-end manner. Specifically, the extracted features are encoded conditioned on a prediction from past quantized latent frames so that temporal correlations are further removed. Moreover, we introduce a learnable compression on the time-frequency input to adaptively adjust the attention paid to main frequencies and details at different bitrates. A differentiable vector quantization scheme based on distance-to-soft mapping and Gumbel-Softmax is proposed to better model the latent distributions with rate constraint. Subjective results on multilingual speech datasets show that, with low latency, the proposed TF-Codec at 1 kbps achieves significantly better quality than Opus at 9 kbps, and TF-Codec at 3 kbps outperforms both EVS at 9.6 kbps and Opus at 12 kbps. Numerous studies are conducted to demonstrate the effectiveness of these techniques.

Driven by a wide range of essential applications, significant achievements have recently been made to explore WiFi-based Human Activity Recognition (HAR) techniques that utilize the information collected by commercial off-the-shelf (COTS) WiFi infrastructures to infer human activities without the need for the subject to carry any devices. Although existing WiFi-based HAR systems achieve satisfactory performance in some instances, they are faced with a severe challenge that the impacts of ubiquitous Co-channel Interference (CCI) on WiFi signals are inevitable. This downgrades the performance of these HAR systems significantly. To address this challenge, we propose PhaseAnti in this paper, a novel WiFi-based HAR system to exploit the CCI-independent phase component, Nonlinear Phase Error Variation (NLPEV), of WiFi Channel State Information (CSI) to cope with the negative effects of CCI. The stability of NLPEV data and the sensibility of this component to motions are rigorously analyzed. Furthermore, validated by extensive properly designed experiments, this phase component across subcarriers is invariant under various CCI scenarios while sufficiently distinct for different motions. Therefore, the NLPEV data can be used and processed effectively to perform HAR in CCI scenarios. Extensive experiments with various daily activities in different indoor rooms demonstrate the superior effectiveness and generalizability of the proposed PhaseAnti system under various CCI scenarios. Specifically, PhaseAnti achieves a$ 96.5\%$recognition accuracy rate (RAR) on average in different CCI scenarios, which can improve up to a$ 16.7\%$RAR compared with the amplitude component in the presence of CCI. Furthermore, the recognition speed is 10.3 × faster than the state-of-the-art solution.

Instance segmentation is a challenging task aiming at classifying and segmenting all object instances of specific classes. While two-stage box-based methods achieve top performances in the image domain, they cannot easily extend their superiority into the video domain. This is because they usually deal with features or images cropped from the detected bounding boxes without alignment, failing to capture pixel-level temporal consistency. We embrace the observation that bottom-up methods dealing with box-free features could offer accurate spacial correlations across frames, which can be fully utilized for object and pixel level tracking. We first propose our bottom-up framework equipped with a temporal context fusion module to better encode inter-frame correlations. Intra-frame cues for semantic segmentation and object localization are simultaneously extracted and reconstructed by corresponding decoders after a shared backbone. For efficient and robust tracking among instances, we introduce an instance-level correspondence across adjacent frames, which is represented by a center-to-center flow, termed as instance flow, to assemble messy dense temporal correspondences. Experiments demonstrate that the proposed method outperforms the state-of-the-art online methods (taking image-level input) on the challenging Youtube-VIS dataset (Yang et al., 2019).

Applying deep learning to video compression has attracted increasing attention in recent few years. In this work, we address end-to-end learned video compression with a special focus on better learning and utilizing temporal contexts. We propose to propagate not only the last reconstructed frame but also the feature before obtaining the reconstructed frame for temporal context mining. From the propagated feature, we learn multi-scale temporal contexts and re-fill the learned temporal contexts into the modules of our compression scheme, including the contextual encoder-decoder, the frame generator, and the temporal context encoder. We discard the parallelization-unfriendly auto-regressive entropy model to pursue a more practical encoding and decoding time. Experimental results show that our proposed scheme achieves a higher compression ratio than the existing learned video codecs. Our scheme also outperforms x264 and x265 (representing industrial software for H.264 and H.265, respectively) as well as the official reference software for H.264, H.265, and H.266 (JM, HM, and VTM, respectively). Specifically, when intra period is 32 and oriented to PSNR, our scheme outperforms H.265–HM by 14.4% bit rate saving; when oriented to MS-SSIM, our scheme outperforms H.266–VTM by 21.1% bit rate saving.

We aim to super-resolve text images from unrecognizable low-resolution inputs. Existing super-resolution methods mainly learn a direct mapping from low-resolution to high-resolution images by exploring low-level features, which usually generate blurry outputs and suffer from severe structure distortion for text parts, especially when the resolution is quite low. Both the visual quality and the readability will suffer. To tackle these issues, we propose a new text super-resolution paradigm by recovering with understanding. Specifically, we extract a text-embedding prior and a text-structure prior from the upsampled image by learning to understand the text. The two priors with rich structure information and text-embedding information are then used as auxiliary information to recover the clear text structure. In addition, we introduce a text-feature loss to guide the training for better text recognizability. Extensive evaluations on both screen and scene text image datasets show that our method largely outperforms the state-of-the-art in both visual quality and recognition accuracy.

Recently, transformer-based image segmentation methods have achieved notable success against previous solutions. While for video domains, how to effectively model temporal context with the attention of object instances across frames remains an open problem. In this paper, we propose an online video instance segmentation framework with a novel instance-aware temporal fusion method. We first leverage the representation, \ie, a latent code in the global context (instance code) and CNN feature maps to represent instance- and pixel-level features. Based on this representation, we introduce a cropping-free temporal fusion approach to model the temporal consistency between video frames. Specifically, we encode global instance-specific information in the instance code and build up inter-frame contextual fusion with hybrid attentions between the instance codes and CNN feature maps. Inter-frame consistency between the instance codes is further enforced with order constraints. By leveraging the learned hybrid temporal consistency, we are able to directly retrieve and maintain instance identities across frames, eliminating the complicated frame-wise instance matching in prior methods. Extensive experiments have been conducted on popular VIS datasets, i.e. Youtube-VIS-19/21. Our model achieves the best performance among all online VIS methods. Notably, our model also eclipses all offline methods when using the ResNet-50 backbone.

Error propagation is a general but crucial problem in online semi-supervised video object segmentation. We aim to suppress error propagation through a correction mechanism with high reliability. The key insight is to disentangle the correction from the conventional mask propagation process with reliable cues. We introduce two modulators, propagation and correction modulators, to separately perform channel-wise recalibration on the target frame embeddings according to local temporal correlations and reliable references respectively. Specifically, we assemble the modulators with a cascaded propagation-correction scheme. This avoids overriding the effects of the reliable correction modulator by the propagation modulator. Although the reference frame with the ground truth label provides reliable cues, it could be very different from the target frame and introduce uncertain or incomplete correlations. We augment the reference cues by supplementing reliable feature patches to a maintained pool, thus offering more comprehensive and expressive object representations to the modulators. In addition, a reliability filter is designed to retrieve reliable patches and pass them in subsequent frames. Our model achieves state-of-the-art performance on YouTube-VOS18, YouTube-VOS19 and DAVIS17-Val/Test benchmarks. Extensive experiments demonstrate that the correction mechanism provides considerable performance gain by fully utilizing reliable guidance.

How to efficiently utilize the temporal features is crucial, yet challenging, for video restoration. The temporal features usually contain various noisy and uncorrelated information, and they may interfere with the restoration of the current frame. This paper proposes learning noiserobust feature representations to help video restoration. We are inspired by that the neural codec is a natural denoiser: In neural codec, the noisy and uncorrelated contents which are hard to predict but cost lots of bits are more inclined to be discarded for bitrate saving. Therefore, we design a neural compression module to filter the noise and keep the most useful information in features for video restoration. To achieve robustness to noise, our compression module adopts a spatial-channel-wise quantization mechanism to adaptively determine the quantization step size for each position in the latent. Experiments show that our method can significantly boost the performance on video denoising, where we obtain 0.13 dB improvement over BasicVSR++ with only 0.23x FLOPs. Meanwhile, our method also obtains SOTA results on video deraining and dehazing.

Contrastive learning between different views of the data achieves outstanding success in the field of self-supervised representation learning and the learned representations are useful in broad downstream tasks. Since all supervision information for one view comes from the other view, contrastive learning approximately obtains the minimal sufficient representation which contains the shared information and eliminates the non-shared information between views. Considering the diversity of the downstream tasks, it cannot be guaranteed that all task-relevant information is shared between views. Therefore, we assume the non-shared task-relevant information cannot be ignored and theoretically prove that the minimal sufficient representation in contrastive learning is not sufficient for the downstream tasks, which causes performance degradation. This reveals a new problem that the contrastive learning models have the risk of overfitting to the shared information between views. To alleviate this problem, we propose to increase the mutual information between the representation and input as regularization to approximately introduce more task-relevant information, since we cannot utilize any downstream task information during training. Extensive experiments verify the rationality of our analysis and the effectiveness of our method. It significantly improves the performance of several classic contrastive learning models in downstream tasks. Our code is available at https://github.com/Haoqing-Wang/InfoCL.

One-shot object detection aims at detecting novel objects according to merely one given instance. With extreme data scarcity, current approaches explore various feature fusions to obtain directly transferable meta-knowledge. Yet, their performances are often unsatisfactory. In this paper, we attribute this to inappropriate correlation methods that misalign query-support semantics by overlooking spatial structures and scale variances. Upon analysis, we leverage the attention mechanism and propose a simple but effective architecture named Semantic-aligned Fusion Transformer (SaFT) to resolve these issues. Specifically, we equip SaFT with a vertical fusion module (VFM) for cross-scale semantic enhancement and a horizontal fusion module (HFM) for cross-sample feature fusion. Together, they broaden the vision for each feature point from the support to a whole augmented feature pyramid from the query, facilitating semantic-aligned associations. Extensive experiments on multiple benchmarks demonstrate the superiority of our framework. Without fine-tuning on novel classes, it brings significant performance gains to one-stage baselines, lifting state-of-the-art results to a higher level.

Deep-learning based methods have shown their advantages in audio coding over traditional ones but limited attention has been paid on real-time communications (RTC). This paper proposes the TFNet, an end-to-end neural speech codec with low latency for RTC. It takes an encoder-temporal filtering-decoder paradigm that has seldom been investigated in audio coding. An interleaved structure is proposed for temporal filtering to capture both short-term and long-term temporal dependencies. Furthermore, with end-to-end optimization, the TFNet is jointly optimized with speech enhancement and packet loss concealment, yielding a one-for-all network for three tasks. Both subjective and objective results demonstrate the efficiency of the proposed TFNet.

Bitrate scalability is a desirable feature for audio coding in real-time communications.Existing neural audio codecs usually enforce a specific bitrate during training, so different models need to be trained for each target bitrate, which increases the memory footprint at the sender and the receiver side and transcoding is often needed to support multiple receivers.In this paper, we introduce a cross-scale scalable vector quantization scheme (CSVQ), in which multi-scale features are encoded progressively with stepwise feature fusion and refinement.In this way, a coarse-level signal is reconstructed if only a portion of the bitstream is received, and progressively improves the quality as more bits are available.The proposed CSVQ scheme can be flexibly applied to any neural audio coding network with a mirrored auto-encoder structure to achieve bitrate scalability.Subjective results show that the proposed scheme outperforms the classical residual VQ (RVQ) with scalability.Moreover, the proposed CSVQ at 3 kbps outperforms Opus at 9 kbps and Lyra at 3kbps and it could provide a graceful quality boost with bitrate increase.

In this paper we propose a multi-modal multi-correlation learning framework targeting at the task of audio-visual speech separation.Although previous efforts have been extensively put on combining audio and visual modalities, most of them solely adopt a straightforward concatenation of audio and visual features.To exploit the real useful information behind these two modalities, we define two key correlations which are: (1) identity correlation (between timbre and facial attributes); (2) phonetic correlation (between phoneme and lip motion).These two correlations together comprise the complete information, which shows a certain superiority in separating target speaker's voice especially in some hard cases, such as the same gender or similar content.For implementation, contrastive learning or adversarial training approach is applied to maximize these two correlations.Both of them work well, while adversarial training shows its advantage by avoiding some limitations of contrastive learning.Compared with previous research, our solution demonstrates clear improvement on experimental metrics without additional complexity.Further analysis reveals the validity of the proposed architecture and its good potential for future extension.

Neural audio coding has shown very promising results recently in the literature to largely outperform traditional codecs but limited attention has been paid on its error resilience.Neural codecs trained considering only source coding tend to be extremely sensitive to channel noises, especially in wireless channels with high error rate.In this paper, we investigate how to elevate the error resilience of neural audio codecs for packet losses that often occur during real-time communications.We propose a feature-domain packet loss concealment algorithm (FD-PLC) for real-time neural speech coding.Specifically, we introduce a self-attention-based module on the received latent features to recover lost frames in the feature domain before the decoder.A hybrid segment-level and frame-level frequencydomain discriminator is employed to guide the network to focus on both the generative quality of lost frames and the continuity with neighbouring frames.Experimental results on several error patterns show that the proposed scheme can achieve better robustness compared with the corresponding error-free and error-resilient baselines.We also show that feature-domain concealment is superior to waveform-domain counterpart as postprocessing.

Video frame synthesis, which consists of interpolation and extrapolation, is an essential video processing technique that can be applied to various scenarios. However, most existing methods cannot handle small objects or large motion well, especially in high-resolution videos such as 4K videos. To eliminate such limitations, we introduce a neighbor correspondence matching (NCM) algorithm for flow-based frame synthesis. Since the current frame is not available in video frame synthesis, NCM is performed in a current-frame-agnostic fashion to establish multi-scale correspondences in the spatial-temporal neighborhoods of each pixel. Based on the powerful motion representation capability of NCM, we further propose to estimate intermediate flows for frame synthesis in a heterogeneous coarse-to-fine scheme. Specifically, the coarse-scale module is designed to leverage neighbor correspondences to capture large motion, while the fine-scale module is more computationally efficient to speed up the estimation process. Both modules are trained progressively to eliminate the resolution gap between training dataset and real-world videos. Experimental results show that NCM achieves state-of-the-art performance on several benchmarks. In addition, NCM can be applied to various practical scenarios such as video compression to achieve better performance.

For neural video codec, it is critical, yet challenging, to design an efficient entropy model which can accurately predict the probability distribution of the quantized latent representation. However, most existing video codecs directly use the ready-made entropy model from image codec to encode the residual or motion, and do not fully leverage the spatial-temporal characteristics in video. To this end, this paper proposes a powerful entropy model which efficiently captures both spatial and temporal dependencies. In particular, we introduce the latent prior which exploits the correlation among the latent representation to squeeze the temporal redundancy. Meanwhile, the dual spatial prior is proposed to reduce the spatial redundancy in a parallel-friendly manner. In addition, our entropy model is also versatile. Besides estimating the probability distribution, our entropy model also generates the quantization step at spatial-channel-wise. This content-adaptive quantization mechanism not only helps our codec achieve the smooth rate adjustment in single model but also improves the final rate-distortion performance by dynamic bit allocation. Experimental results show that, powered by the proposed entropy model, our neural codec can achieve 18.2% bitrate saving on UVG dataset when compared with H.266 (VTM) using the highest compression ratio configuration. It makes a new milestone in the development of neural video codec. The codes are at https://github.com/microsoft/DCVC.

In the booming video era, video segmentation attracts increasing research attention in the multimedia community. Semi-supervised video object segmentation (VOS) aims at segmenting objects in all target frames of a video, given annotated object masks of reference frames. Most existing methods build pixel-wise reference-target correlations and then perform pixel-wise tracking to obtain target masks. Due to neglecting object-level cues, pixel-level approaches make the tracking vulnerable to perturbations, and even indiscriminate among similar objects. Towards robust VOS, the key insight is to calibrate the representation and mask of each specific object to be expressive and discriminative. Accordingly, we propose a new deep network, which can adaptively construct object representations and calibrate object masks to achieve stronger robustness. First, we construct the object representations by applying an adaptive object proxy (AOP) aggregation method, where the proxies represent arbitrary-shaped segments at multi-levels for reference. Then, prototype masks are initially generated from the reference-target correlations based on AOP. Afterwards, such proto-masks are further calibrated through network modulation, conditioning on the object proxy representations. We consolidate this conditional mask calibration process in a progressive manner, where the object representations and proto-masks evolve to be discriminative iteratively. Extensive experiments are conducted on the standard VOS benchmarks, YouTube-VOS-18/19 and DAVIS-17. Our model achieves the state-of-the-art performance among existing published works, and also exhibits superior robustness against perturbations.

Obtaining the human-like perception ability of abstracting visual concepts from concrete pixels has always been a fundamental and important target in machine learning research fields such as disentangled representation learning and scene decomposition. Towards this goal, we propose an unsupervised transformer-based Visual Concepts Tokenization framework, dubbed VCT, to perceive an image into a set of disentangled visual concept tokens, with each concept token responding to one type of independent visual concept. Particularly, to obtain these concept tokens, we only use cross-attention to extract visual information from the image tokens layer by layer without self-attention between concept tokens, preventing information leakage across concept tokens. We further propose a Concept Disentangling Loss to facilitate that different concept tokens represent independent visual concepts. The cross-attention and disentangling loss play the role of induction and mutual exclusion for the concept tokens, respectively. Extensive experiments on several popular datasets verify the effectiveness of VCT on the tasks of disentangled representation learning and scene decomposition. VCT achieves the state of the art results by a large margin.

For deep reinforcement learning (RL) from pixels, learning effective state representations is crucial for achieving high performance. However, in practice, limited experience and high-dimensional inputs prevent effective representation learning. To address this, motivated by the success of mask-based modeling in other research fields, we introduce mask-based reconstruction to promote state representation learning in RL. Specifically, we propose a simple yet effective self-supervised method, Mask-based Latent Reconstruction (MLR), to predict complete state representations in the latent space from the observations with spatially and temporally masked pixels. MLR enables better use of context information when learning state representations to make them more informative, which facilitates the training of RL agents. Extensive experiments show that our MLR significantly improves the sample efficiency in RL and outperforms the state-of-the-art sample-efficient RL methods on multiple continuous and discrete control benchmarks. Our code is available at https://github.com/microsoft/Mask-based-Latent-Reconstruction.

Temporal modeling is crucial for various video learning tasks. Most recent approaches employ either factorized (2D+1D) or joint (3D) spatial-temporal operations to extract temporal contexts from the input frames. While the former is more efficient in computation, the latter often obtains better performance. In this paper, we attribute this to a dilemma between the sufficiency and the efficiency of interactions among various positions in different frames. These interactions affect the extraction of task-relevant information shared among frames. To resolve this issue, we prove that frame-by-frame alignments have the potential to increase the mutual information between frame representations, thereby including more task-relevant information to boost effectiveness. Then we propose Alignment-guided Temporal Attention (ATA) to extend 1-dimensional temporal attention with parameter-free patch-level alignments between neighboring frames. It can act as a general plug-in for image backbones to conduct the action recognition task without any model-specific design. Extensive experiments on multiple benchmarks demonstrate the superiority and generality of our module.

Detecting and localizing objects in the real 3D space, which plays a crucial role in scene understanding, is particularly challenging given only a monocular image due to the geometric information loss during imagery projection. We propose MonoGRNet for the amodal 3D object detection from a monocular image via geometric reasoning in both the observed 2D projection and the unobserved depth dimension. MonoGRNet decomposes the monocular 3D object detection task into four sub-tasks including 2D object detection, instance-level depth estimation, projected 3D center estimation and local corner regression. The task decomposition significantly facilitates the monocular 3D object detection, allowing the target 3D bounding boxes to be efficiently predicted in a single forward pass, without using object proposals, post-processing or the computationally expensive pixel-level depth estimation utilized by previous methods. In addition, MonoGRNet flexibly adapts to both fully and weakly supervised learning, which improves the feasibility of our framework in diverse settings. Experiments are conducted on KITTI, Cityscapes and MS COCO datasets. Results demonstrate the promising performance of our framework in various scenarios.

Time-variant factors including dynamic delay and varying echo path often occur in real-world acoustic echo cancellation (AEC) applications. Current end-to-end deep neural network (DNN) based methods usually model the time-variant components implicitly and can hardly handle the unpredictable time-variance in real-time AEC. To explicitly capture the time-variant components, we propose a dynamic kernel generation (DKG) module that can be introduced as a learnable plug-in to a DNN-based end-to-end pipeline. Specifically, the DKG module generates a convolutional kernel regarding to each input audio frame, so that the DNN model is able to dynamically adjust its weights according to the input signal during inference. Experimental results verify that DKG module improves the AEC performance of the model under time-variant scenarios, especially in the double-talk cases.

Weakly-supervised temporal action localization aims to learn detecting temporal intervals of action classes with only video-level labels. To this end, it is crucial to separate frames of action classes from the background frames (i.e., frames not belonging to any action classes). In this paper, we present a new perspective on background frames where they are modeled as out-of-distribution samples regarding their inconsistency. Then, background frames can be detected by estimating the probability of each frame being out-of-distribution, known as uncertainty, but it is infeasible to directly learn uncertainty without frame-level labels. To realize the uncertainty learning in the weakly-supervised setting, we leverage the multiple instance learning formulation. Moreover, we further introduce a background entropy loss to better discriminate background frames by encouraging their in-distribution (action) probabilities to be uniformly distributed over all action classes. Experimental results show that our uncertainty modeling is effective at alleviating the interference of background frames and brings a large performance gain without bells and whistles. We demonstrate that our model significantly outperforms state-of-the-art methods on the benchmarks, THUMOS'14 and ActivityNet (1.2 & 1.3). Our code is available at https://github.com/Pilhyeon/WTAL-Uncertainty-Modeling.

Visible-infrared cross modality person re-identification (VI-ReID) is a core but challenging technology in the 24-hours intelligent surveillance system. How to eliminate the large modality gap lies in the heart of VI-ReID. Conventional methods mainly focus on directly aligning the heterogeneous modalities into the same space. However, due to the unbalanced color information between the visible and infrared images, the features of visible images tend to overfit the clothing color information, which would be harmful to the modality alignment. Besides, these methods mainly align the heterogeneous feature distributions in dataset-level while ignoring the valuable identity information, which may cause the feature misalignment of some identities and weaken the discrimination of features. To tackle above problems, we propose a novel approach for VI-ReID. It learns the color-irrelevant features through the color-irrelevant consistency learning (CICL) and aligns the identity-level feature distributions by the identity-aware modality adaptation (IAMA). The CICL and IAMA are integrated into a joint learning framework and can promote each other. Extensive experiments on two popular datasets SYSU-MM01 and RegDB demonstrate the superiority and effectiveness of our approach against the state-of-the-art methods.

Speech enhancement is challenging because of the diversity of background noise types. Most of the existing methods are focused on modelling the speech rather than the noise. In this paper, we propose a novel idea to model speech and noise simultaneously in a two-branch convolutional neural network, namely SN-Net. In SN-Net, the two branches predict speech and noise, respectively. Instead of information fusion only at the final output layer, interaction modules are introduced at several intermediate feature domains between the two branches to benefit each other. Such an interaction can leverage features learned from one branch to counteract the undesired part and restore the missing component of the other and thus enhance their discrimination capabilities. We also design a feature extraction module, namely residual-convolution-and-attention (RA), to capture the correlations along temporal and frequency dimensions for both the speech and the noises. Evaluations on public datasets show that the interaction module plays a key role in simultaneous modeling and the SN-Net outperforms the state-of-the-art by a large margin on various evaluation metrics. The proposed SN-Net also shows superior performance for speaker separation.

Self-attention has been successfully applied to video representation learning due to the effectiveness of modeling long range dependencies. Existing approaches build the dependencies merely by computing the pairwise correlations along spatial and temporal dimensions simultaneously. However, spatial correlations and temporal correlations represent different contextual information of scenes and temporal reasoning. Intuitively, learning spatial contextual information first will benefit temporal modeling. In this paper, we propose a separable self-attention (SSA) module, which models spatial and temporal correlations sequentially, so that spatial contexts can be efficiently used in temporal modeling. By adding SSA module into 2D CNN, we build a SSA network (SSAN) for video representation learning. On the task of video action recognition, our approach outperforms state-of-the-art methods on Something-Something and Kinetics-400 datasets. Our models often outperform counterparts with shallower network and fewer modalities. We further verify the semantic learning ability of our method in visual-language task of video retrieval, which showcases the homogeneity of video representations and text embeddings. On MSR-VTT and Youcook2 datasets, video representations learnt by SSA significantly improve the state-of-the-art performance.

Existing speech enhancement methods mainly separate speech from noises at the signal level or in the time-frequency domain. They seldom pay attention to the semantic information of a corrupted signal. In this paper, we aim to bridge this gap by extracting phoneme identities to help speech enhancement. Specifically, we propose a phoneme-based distribution regularization (PbDr) for speech enhancement, which incorporates frame-wise phoneme information into speech enhancement network in a conditional manner. As different phonemes always lead to different feature distributions in frequency, we propose to learn a parameter pair, i.e. scale and bias, through a phoneme classification vector to modulate the speech enhancement network. The modulation parameter pair includes not only frame-wise but also frequency-wise conditions, which effectively map features to phoneme-related distributions. In this way, we explicitly regularize speech enhancement features by recognition vectors. Experiments on public datasets demonstrate that the proposed PbDr module can not only boost the perceptual quality for speech enhancement but also the recognition accuracy of an ASR system on the enhanced speech. This PbDr module could be readily incorporated into other speech enhancement networks as well.

Self-supervised learning has been successfully applied to pre-train video representations, which aims at efficient adaptation from pre-training domain to downstream tasks. Existing approaches merely leverage contrastive loss to learn instance-level discrimination. However, lack of category information will lead to hard-positive problem that constrains the generalization ability of this kind of methods. We find that the multi-task process of meta learning can provide a solution to this problem. In this paper, we propose a Meta-Contrastive Network (MCN), which combines the contrastive learning and meta learning, to enhance the learning ability of existing self-supervised approaches. Our method contains two training stages based on model-agnostic meta learning (MAML), each of which consists of a contrastive branch and a meta branch. Extensive evaluations demonstrate the effectiveness of our method. For two downstream tasks, i.e., video action recognition and video retrieval, MCN outperforms state-of-the-art approaches on UCF101 and HMDB51 datasets. To be more specific, with R(2+1)D backbone, MCN achieves Top-1 accuracies of 84.8% and 54.5% for video action recognition, as well as 52.5% and 23.7% for video retrieval.

Geometry Projection is a powerful depth estimation method in monocular 3D object detection. It estimates depth dependent on heights, which introduces mathematical priors into the deep model. But projection process also introduces the error amplification problem, in which the error of the estimated height will be amplified and reflected greatly at the output depth. This property leads to uncontrollable depth inferences and also damages the training efficiency. In this paper, we propose a Geometry Uncertainty Projection Network (GUP Net) to tackle the error amplification problem at both inference and training stages. Specifically, a GUP module is proposed to obtains the geometry-guided uncertainty of the inferred depth, which not only provides high reliable confidence for each depth but also benefits depth learning. Furthermore, at the training stage, we propose a Hierarchical Task Learning strategy to reduce the instability caused by error amplification. This learning algorithm monitors the learning situation of each task by a proposed indicator and adaptively assigns the proper loss weights for different tasks according to their pre-tasks situation. Based on that, each task starts learning only when its pre-tasks are learned well, which can significantly improve the stability and efficiency of the training process. Extensive experiments demonstrate the effectiveness of the proposed method. The overall model can infer more reliable object depth than existing methods and outperforms the state-of-the-art image-based monocular 3D detectors by 3.74% and 4.7% AP40of the car and pedestrian categories on the KITTI benchmark. The code and model will be released at https://github.com/SuperMHP/GUPNet.

User identity authentication is essencial for ensuring information security. With the widespread use of electronic devices, online handwriting verification becomes more important in identity authentication based on biometrics and widely used in financial, commercial, and forensic fields. In this paper, we propose a multi-path feature fusion network for multi-modal fusion of static and dynamic handwriting obtained by electronic devices to intensify the handwriting verification. Since traditional handwritten signature verification, of which the handwritten content just the writer’s name, is vulnerable to skilled forgery attacks, we propose a content-independent handwriting verification scheme to solve this problem. We also build a handwriting dataset with approximately 5400 samples of 30 individuals’ handwriting, which contributes to extracting content-independent handwriting style features. We test our method on widely used BiosecurID dataset and our dataset. The experimental results demonstrate the feasibility of the proposed method.

Most of the existing neural video compression methods adopt the predictive coding framework, which first generates the predicted frame and then encodes its residue with the current frame. However, as for compression ratio, predictive coding is only a sub-optimal solution as it uses simple subtraction operation to remove the redundancy across frames. In this paper, we propose a deep contextual video compression framework to enable a paradigm shift from predictive coding to conditional coding. In particular, we try to answer the following questions: how to define, use, and learn condition under a deep video compression framework. To tap the potential of conditional coding, we propose using feature domain context as condition. This enables us to leverage the high dimension context to carry rich information to both the encoder and the decoder, which helps reconstruct the high-frequency contents for higher video quality. Our framework is also extensible, in which the condition can be flexibly designed. Experiments show that our method can significantly outperform the previous state-of-the-art (SOTA) deep video compression methods. When compared with x265 using veryslow preset, we can achieve 26.0% bitrate saving for 1080P standard test videos.

For widely used real-time applications, encoding and transmitting multiple videos jointly over a limited bandwidth has become a popular topic. Allocating different bitrates for different sources is a better way to meet different demands from applications. In this paper, we focus on providing equal quality to users by minimizing the variance of distortion among sequences, which is denoted as the minVAR problem. The state-of-the-art Look-ahead and Feed-back Allocation Model (LFAM) allocates bitrate by taking both look-ahead complexity measures and feed-back information into consideration. However, LFAM brings additional delay to real-time applications. By taking the bitrate allocation problem as a time-series decision making problem, we propose a Deep-Reinforcement-Learning-based approach to allocate bitrate with only feed-back information to solve the two-source minVAR problem. Afterward, we introduce a binary-tree-based hierarchical approach to apply our model to arbitrary number of sources. Tested with the widely used open-source x264 encoder, our approach decreases the variance compared with LFAM in all experiments under two-, three- and four-source scenarios. Furthermore, the proposed approach also outperforms LFAM in the mean quality. The proposed approach is insensitive to the order of sequences and encoders with different complexities, showing its robustness and generalization capability.

As an important topic in the multimedia and computer vision fields, salient object detection has been researched for years. Recently, state-of-the-art performance has been witnessed with the aid of the fully convolutional networks (FCNs) and the various pyramid-like encoder-decoder frameworks. Starting from a common encoder-decoder architecture, we enhance a residual refinement network with feature purification for better saliency estimation. To this end, we improve the global knowledge streams with intermediate supervisions for global saliency estimation and design a specific feature subtraction module for residual learning, respectively. On the basis of the strengthened network, we also introduce an attribute encoding sub-network (AENet) with a grid aggregation block (GAB) to guide the final saliency predictor to obtain more accurate saliency maps. Furthermore, the network is trained with a novel constraint loss besides the traditional cross-entropy loss to yield the finer results. Extensive experiments on five public benchmarks show our method achieves better or comparable performance compared with previous state-of-the-art methods.

Affinity, which represents whether two pixels belong to a same instance, is an equivalent representation to the instance segmentation labels. Conventional works do not make an explicit exploration on the affinity. In this article, we present two instance segmentation schemes based on pixel affinity information and show the effectiveness of affinity in both aspects. For proposal-free method, we predict pixel affinity for each image and then propose a simple yet effective graph merge algorithm to cluster pixels into instances. It shows that the affinity is powerful as an instance-relevant information to guide the clustering procedure in proposal-free instance segmentation. For proposal-based methods, we extend conventional framework with affinity head and introduce affinity as attached supervision in training phase. Without any additional inference cost, we can improve the performance of existing proposal-based instance segmentation methods, which shows that the affinity can also be applied as an auxiliary loss and training with such extra loss is beneficial to the training progress. Experimental results show that our schemes achieve comparable performance to other state-of-the-art instance segmentation methods. With Cityscapes training data, the proposed proposal-free method achieves 28.8 AP and the proposal-based method gets 27.2 AP both on test sets.

Online learning is crucial to robust visual object tracking as it can provide high discrimination power in the presence of background distractors. However, there are two contradictory factors affecting its successful deployment on the real visual tracking platform: the discrimination issue due to the challenges in vanilla gradient descent, which does not guarantee good convergence; the robustness issue due to over-fitting resulting from excessive update with limited memory size (the oldest samples are discarded). Despite many dedicated techniques proposed to somehow treat those issues, in this paper we take a new way to strike a compromise between them based on the recursive least-squares estimation (LSE) algorithm. After connecting each fully-connected layer with LSE separately via normal equations, we further propose an improved mini-batch stochastic gradient descent algorithm for fully-connected network learning with memory retention in a recursive fashion. This characteristic can spontaneously reduce the risk of over-fitting resulting from catastrophic forgetting in excessive online learning. Meanwhile, it can effectively improve convergence though the cost function is computed over all the training samples that the algorithm has ever seen. We realize this recursive LSE-aided online learning technique in the state-of-the-art RT-MDNet tracker, and the consistent improvements on four challenging benchmarks prove its efficiency without additional offline training and too much tedious work on parameter adjusting.

Cross-modality person re-identification (cm-ReID) is a challenging but key technology for intelligent video analysis. Existing works mainly focus on learning modality-shared representation by embedding different modalities into a same feature space, lowering the upper bound of feature distinctiveness. In this paper, we tackle the above limitation by proposing a novel cross-modality shared-specific feature transfer algorithm (termed cm-SSFT) to explore the potential of both the modality-shared information and the modality-specific characteristics to boost the reidentification performance. We model the affinities of different modality samples according to the shared features and then transfer both shared and specific features among and across modalities. We also propose a complementary feature learning strategy including modality adaption, project adversarial learning and reconstruction enhancement to learn discriminative and complementary shared and specific features of each modality, respectively. The entire cmSSFTalgorithm can be trained in an end-to-end manner. We conducted comprehensive experiments to validate the superiority ofthe overall algorithm and the effectiveness ofeach component. The proposed algorithm significantly outperforms state-of-the-arts by 22.5% and 19.3% mAP on the two mainstream benchmark datasets SYSU-MM01 and RegDB, respectively.

A crucial task in scene understanding is 3D object detection, which aims to detect and localize the 3D bounding boxes of objects belonging to specific classes. Existing 3D object detectors heavily rely on annotated 3D bounding boxes during training, while these annotations could be expensive to obtain and only accessible in limited scenarios. Weakly supervised learning is a promising approach to reducing the annotation requirement, but existing weakly supervised object detectors are mostly for 2D detection rather than 3D. In this work, we propose VS3D, a framework for weakly supervised 3D object detection from point clouds without using any ground truth 3D bounding box for training. First, we introduce an unsupervised 3D proposal module that generates object proposals by leveraging normalized point cloud densities. Second, we present a cross-modal knowledge distillation strategy, where a convolutional neural network learns to predict the final results from the 3D object proposals by querying a teacher network pretrained on image datasets. Comprehensive experiments on the challenging KITTI dataset demonstrate the superior performance of our VS3D in diverse evaluation settings. The source code and pretrained models are publicly available at https://github.com/Zengyi-Qin/Weakly-Supervised-3D-Object-Detection.

Real-time video enhancement is in great demand due to the extensive usage of live video applications, but existing approaches are far from satisfying the strict requirements of speed and stability. We present a novel convolutional network that can perform high-quality enhancement on 1080p videos at 45 FPS with a single CPU, which has high potential for real-world deployment. The proposed network is designed based on a light-weight image network and further consolidated for temporal consistency with a temporal feature aggregation (TFA) module. Unlike most image translation networks that use decoders to generate target images, our network discards decoders and employs only an encoder and a small head. The network predicts color mapping functions instead of pixel values in a grid-like container which fits the CNN structure well and also advances the enhancement to be scalable to any video resolution. Furthermore, the temporal consistency of the output will be enforced by the TFA module which utilizes the learned temporal coherence of semantics across frames. We also demonstrate that the mapping representation is general to various enhancement tasks, such as relighting, retouching and dehazing, on benchmark datasets. Our approach achieves the state-of-the-art performance and performs about 10 times faster than the current real-time method on high-resolution videos.

Albeit the highest accuracy of object detection is generally acquired by multi-stage detectors, like R-CNN and its extension approaches, the single-stage object detectors also achieve remarkable performance with faster execution and higher scalability. Inspired by this, we propose a single-stage framework to tackle the instance segmentation task. Building on a single-stage object detection network in hand, our model outputs the detected bounding box of each instance, the semantic segmentation result, and the pixel affinity simultaneously. After that, we generate the final instance masks via a fast post-processing method with the help of the three outputs above. As far as we know, it is the first attempt to segment instances in a single-stage pipeline on challenging datasets. Extensive experiments demonstrate the efficiency of our post-processing method, and the proposed framework obtains competitive results as a single-stage instance segmentation method. We achieve 32.5 box AP and 26.0 mask AP on the COCO validation set with 500 pixels input scale and 22.9 mask AP on the Cityscapes test set.

Localizing objects in the real 3D space, which plays a crucial role in scene understanding, is particularly challenging given only a single RGB image due to the geometric information loss during imagery projection. We propose MonoGRNet for the amodal 3D object localization from a monocular RGB image via geometric reasoning in both the observed 2D projection and the unobserved depth dimension. MonoGRNet is a single, unified network composed of four task-specific subnetworks, responsible for 2D object detection, instance depth estimation (IDE), 3D localization and local corner regression. Unlike the pixel-level depth estimation that needs per-pixel annotations, we propose a novel IDE method that directly predicts the depth of the targeting 3D bounding box’s center using sparse supervision. The 3D localization is further achieved by estimating the position in the horizontal and vertical dimensions. Finally, MonoGRNet is jointly learned by optimizing the locations and poses of the 3D bounding boxes in the global context. We demonstrate that MonoGRNet achieves state-of-the-art performance on challenging datasets.

In this paper, we address the problem of reconstructing an object’s surface from a single image using generative networks. First, we represent a 3D surface with an aggregation of dense point clouds from multiple views. Each point cloud is embedded in a regular 2D grid aligned on an image plane of a viewpoint, making the point cloud convolution-favored and ordered so as to fit into deep network architectures. The point clouds can be easily triangulated by exploiting connectivities of the 2D grids to form mesh-based surfaces. Second, we propose an encoder-decoder network that generates such kind of multiple view-dependent point clouds from a single image by regressing their 3D coordinates and visibilities. We also introduce a novel geometric loss that is able to interpret discrepancy over 3D surfaces as opposed to 2D projective planes, resorting to the surface discretization on the constructed meshes. We demonstrate that the multi-view point regression network outperforms state-of-the-art methods with a significant improvement on challenging datasets.

In this paper, we study the problem of 3D object detection from stereo images, in which the key challenge is how to effectively utilize stereo information. Different from previous methods using pixel-level depth maps, we propose to employ 3D anchors to explicitly construct object-level correspondences between the regions of interest in stereo images, from which the deep neural network learns to detect and triangulate the targeted object in 3D space. We also introduce a cost-efficient channel reweighting strategy that enhances representational features and weakens noisy signals to facilitate the learning process. All of these are flexibly integrated into a solid baseline detector that inputs monocular images. We demonstrate that both the monocular baseline and the stereo triangulation learning network outperform the prior state-of-the-arts in 3D object detection and localization on the challenging KITTI dataset.

Person Search plays the role of the ultimate destination of person re-identification (re-ID) in real applications. It has many challenges that person re-ID doesn't need to handle, such as mis-detections, false alarms and multi-scale matching. In contrast to previous works, we show that a strong multi-scale person matching system can result in a good person search performance with a common deep object detector (e.g. Faster-RCNN). In this work, we provide a robust person search method called Dynamic Hierarchical Feature Fusion (DHFF) which is based on multi-level feature fusion to tackle with multi-scale matching. In addition, A Multi-Metric loss is proposed to train the model effectively and stably with numerous identities. We evaluate our method on two large person search benchmark data sets: CUHK-SYSU and PRW. Experiments show that the proposed algorithm outperforms other state-of-the-art person search methods.

Establishing an interactive screen sharing system that supports ultra high resolution (such as 4k) is challenging, with latency and frame rate playing important roles in user experience. The screen frame needs to be compressed efficiently without consuming extensive computational resources. We present a hardware-accelerated system for real-time screen sharing, which decreases encoding workload by exploiting content redundancies between successive screen frames. We propose a multiple codec approach that utilizes several encoders with H.264 Advanced Video Coding (H.264/AVC) of different input sizes, creating savings in encoding time by selecting the appropriate one for updated screen content. An optimized metadata processing method is proposed as well. Small but distant updates within a frame can be split into independent frames for more efficient compression, which is also beneficial for interactive latency. In the evaluation, the proposed system takes less encoding time than general single codec implementation in common screen sharing scenarios. Measurement for latency shows that the end-to-end latency for 4K resolution screen sharing is only about 17-25 ms, which makes the proposed system suitable for various applications in local wired and wireless connections.

Objects for detection usually have distinct characteristics in different sub-regions and different aspect ratios. However, in prevalent two-stage object detection methods, Region-of-Interest (RoI) features are extracted by RoI pooling with little emphasis on these translation-variant feature components. We present feature selective networks to reform the feature representations of RoIs by exploiting their disparities among sub-regions and aspect ratios. Our network produces the sub-region attention bank and aspect ratio attention bank for the whole image. The RoI-based sub-region attention map and aspect ratio attention map are selectively pooled from the banks, and then used to refine the original RoI features for RoI classification. Equipped with a lightweight detection subnetwork, our network gets a consistent boost in detection performance based on general ConvNet backbones (ResNet-101, GoogLeNet and VGG-16). Without bells and whistles, our detectors equipped with ResNet-101 achieve more than 3% mAP improvement compared to counterparts on PASCAL VOC 2007, PASCAL VOC 2012 and MS COCO datasets.

Screen content coding plays an important role in many applications. To meet the growing demands of screen content coding, the emerging AV1 video codec incorporates several coding tools, which are specially designed for screen content utilizing its distinctive characteristics. Among these tools, the intra block copy utilizes the characteristic that repeating patterns frequently occur in screen content. This paper presents the technology of intra block copy in AV1. In particular, to efficiently search the predictor in the reconstructed regions of the current picture, AV1 uses the hash matching method at the encoder side. For the generation of hash table, a bottom-to-up manner is adopted to reduce the redundant computation and then decrease the encoding time. In addition, several constraints are involved to facilitate hardware design. Experimental results demonstrate that the intra block copy in AV1 can bring 27.1% bitrate saving for screen content. When compared with the non hash-based intra block copy, the hash-based method achieves 12.2% bitrate saving.

Cloud gaming as the emerging game service has attracted significant attention. However, traditional video streaming approach suffers from high bandwidth consumption, and traditional graphics streaming approach requires a long initial period to download game models. In this paper, we propose a novel hybrid streaming framework, jointly applying video streaming and graphics streaming to provide a high-quality gaming experience. In the proposed framework, cloud servers not only transmit the encoded video frames but also progressively transmit the graphics data, which are used to render a game frame to provide an additional reference to the video encoder. Based on the proposed framework, we investigate the delay-rate-distortion optimization problem, where the source rate between the video stream and the graphics stream is optimized to minimize the overall distortion under the bandwidth and response delay constraints. The experimental results demonstrate that the proposed hybrid streaming can achieve the lowest distortion under the constraints of bandwidth and response delay, compared with the traditional video streaming and graphics streaming.

The latest high efficiency video coding (HEVC) standard achieves about 50% bit-rate reduction at equivalent visual quality compared to H.264/AVC. Sample adaptive offset (SAO) is one of the newly adopted tools right after deblocking filter, which can improve both coding efficiency and visual quality. However, for real-time encoding scenarios, the complexity of SAO is usually too high to be enabled. In this paper, a GPU-based optimization algorithm is proposed to reduce the complexity of SAO. Experiments are conducted based on the state-of-the-art open source HEVC encoder, i.e. X265. Results show that the proposed algorithm can reduce about 70% processing time of SAO on average without sacrifice of coding efficiency.

The pervasive computing environment and wide network bandwidth provide users more opportunities to share screen content among multiple devices. In this article, we introduce a remote display system to enable screen sharing among multiple devices with high fidelity and responsive interaction. In the developed system, the frame-level screen content is compressed and transmitted to the client side for screen sharing, and the instant control inputs are simultaneously transmitted to the server side for interaction. Even if the screen responds immediately to the control messages and updates at a high frame rate on the server side, it is difficult to update the screen content with low delay and high frame rate in the client side due to non-negligible time consumption on the whole screen frame compression, transmission, and display buffer updating. To address this critical problem, we propose a layered structure for screen coding and rendering to deliver diverse screen content to the client side with an adaptive frame rate. More specifically, the interaction content with small region screen update is compressed by a blockwise screen codec and rendered at a high frame rate to achieve smooth interaction, while the natural video screen content is compressed by standard video codec and rendered at a regular frame rate for a smooth video display. Experimental results with real applications demonstrate that the proposed system can successfully reduce transmission bandwidth cost and interaction delay during screen sharing. Especially for user interaction in small regions, the proposed system can achieve a higher frame rate than most previous counterparts.

In many multimedia applications, such as online speech, video chat and online conference, multiple source videos are synthesized in a single scene for explicit presentation and the synthesized video is compressed for transmission. The source video with important contents deserves more compression resources for quality preservation under the bandwidth constraint. To address this problem, a region-of-interest (ROI) based coding scheme for synthesized video is proposed in this paper aiming at achieve better and consistent quality for ROI source videos with the bitrate meeting the constraint bandwidth. In the proposed coding scheme, ROI based rate-distortion (R-D) models are established, in which different R-D models are built for different source video. Then an objective function is defined with respect to the video quality and the consistency of video quality. By minimizing the objective function, the optimal quantization parameters for the ROI and non-ROI source videos are obtained. The experimental results show that the proposed coding scheme achieves better and consistent quality for ROI source videos.

Cloud gaming, the newest entry in the online gaming world, leverages the well-known concept of cloud computing to provide real-time gaming services to players. The idea in cloud gaming is to capture the game events from players and transmit them to the cloud, process those events and run the game logic in the cloud, render the game scene as video in the cloud, and stream that video to the players. The advantage is that as long as the client can display video, which pretty much all smartphones, tablets, game consoles, desktops, laptops, and mobile devices today do, the user can play the game without installing it locally, and without needing to have a machine with high-grade 3-D graphics rendering and powerful computing hardware and software. This makes cloud gaming accessible to a huge market of mass consumers. While some variations of cloud gaming systems stream 3-D graphics, in addition to or instead of video, the great majority of cloud gaming implementations are video based. Using the well-known concept of software as a service, cloud gaming is also sometimes referred to as gaming as a service, which is already available as commercial products, such as Sony’s PlayStation Now, Ubitus’s GameNow, G-Cluster, Crytek’s GFACE, PlayGiga, and LiquidSky, to name a few. There are also many efforts concentrating specifically on the underlying technology behind cloud gaming, such as NVIDIA’s Grid, OTOY, CiiNow, Kalydo, and GamingAnywhere [4], the latter being the only open source and free technology. Microsoft is also exploring cloud gaming technologies, with recent successes such as its Kahawai project [1].

With the development of depth data acquisition technologies, access to high-precision depth with more than 8-b depths has become much easier and determining how to efficiently represent and compress high-precision depth is essential for practical depth storage and transmission systems. In this paper, we propose a layered high-precision depth compression framework based on an 8-b image/video encoder to achieve efficient compression with low complexity. Within this framework, considering the characteristics of the high-precision depth, a depth map is partitioned into two layers: 1) the most significant bits (MSBs) layer and 2) the least significant bits (LSBs) layer. The MSBs layer provides rough depth value distribution, while the LSBs layer records the details of the depth value variation. For the MSBs layer, an error-controllable pixel domain encoding scheme is proposed to exploit the data correlation of the general depth information with sharp edges and to guarantee the data format of LSBs layer is 8 b after taking the quantization error from MSBs layer. For the LSBs layer, standard 8-b image/video codec is leveraged to perform the compression. The experimental results demonstrate that the proposed coding scheme can achieve real-time depth compression with satisfactory reconstruction quality. Moreover, the compressed depth data generated from this scheme can achieve better performance in view synthesis and gesture recognition applications compared with the conventional coding schemes because of the error control algorithm.

This paper proposes an efficient content adaptive screen image scaling scheme for the real-time screen applications like remote desktop and screen sharing. In the proposed screen scaling scheme, a screen content classification step is first introduced to classify the screen image into text and pictorial regions. Afterward, we propose an adaptive shift linear interpolation algorithm to predict the new pixel values with the shift offset adapted to the content type of each pixel. The shift offset for each screen content type is offline optimized by minimizing the theoretical interpolation error based on the training samples respectively. The proposed content adaptive screen image scaling scheme can achieve good visual quality and also keep the low complexity for realtime applications.

As the popularity of smart phones and tablets, users have an increasing desire to enjoy ubiquitous game playing. The emerging cloud gaming turns this desire into reality, enabling users to play games at anywhere on any devices. However, due to the huge amount of data transmission, it is challenging to provide a high quality game experience under the limited bandwidth capacity. In this paper, we propose a novel cloud gaming framework, in which we introduce two synchronized graphics buffers at both the server and the client sides. The server not only streams the compressed frames captured from game scenes, but also progressively transmits graphics data. The received graphics data is used to generate reference frames. When compressing the next frame, the cloud server will choose the reference frame with a lower residual error, from the previous frame and the current frame rendered from the graphics buffer. With the accumulation of graphics data, the frame rendered from the graphics buffer is close to the captured frame, which greatly reduces the transmission bit rates. Based on the proposed framework, we study the rate allocation problem, in which we optimize the allocated bit rates between the compressed frame and the graphics data to minimize the total distortion under the bandwidth constraint. Experimental results demonstrate that the proposed framework can optimally allocate bit rates to achieve a minimal distortion for cloud gaming compared to the traditional video streaming and graphics streaming approaches.

The emerging cloud gaming technology has been growing fast, driving up huge mobile consumer demands. The video streaming based cloud gaming scenario renders the game scenes in the cloud servers, and streams the encoded sequences to the thin clints where the game scenes are decoded and displayed to the players. However, current existing clouding gaming services have some problems, such as the latency and bandwidth limitation. The size of the video stream is usually quite large which requires heavy transmission. Worse still, the frame data rate will burst when the game scenes contain fast translation or rotation, resulting in strong latency problem. In this paper, we propose a novel video streaming based cloud gaming algorithm which reduces the burst of the frame rate significantly. There are mainly two innovations in this paper. Firstly, based on the analysis of the motion estimation strategy in the video codec, we introduce image homography technique for better motion prediction. Meanwhile, according to the rasterization rules of the game engine, we present a special designed interpolation algorithm named Edge Preserved Interpolation (EPI), for more accurate edge interpolation and further reduce the residues in the edge regions. The proposed algorithm is implemented on the x264 platform. Experimental results show that our algorithm has 18.0% BD-rate reduction compared with x264.

In this paper, we propose a high frame rate screen video compression scheme aiming at improving the interactive user experience on screen sharing applications. The proposed screen video compression is performed as two-layer coding: a base layer coding using the conventional video codec and an enhancement layer coding using the proposed open-loop coding scheme. For efficient frame level layer selection and compression, the content update of each frame is evaluated through global motion detection. The screen frame with significant content update is fed to the conventional video encoder in base layer. In contrast, the frame with little update is compressed in enhancement layer in which the duplicate content is indicated by global motion vector and skip flag while the updated content is encoded by distinct intra modes in terms of inherent local features. The experimental results demonstrate that for the screen video containing interaction, the proposed coding scheme can achieve 3.09ms/frame encoding rate and 2.33ms/frame decoding rate with efficient rate distortion performance.

Virtual screen system is becoming an essential part in the mobile cloud computing platform. However, designing a low-latency interactive communication for the high-resolution screen content is still challenging due to the network dynamics and the unique characteristics of screen content. In this paper we propose a H.264 based low-latency screen sharing system. To achieve high play-out frame rate, we decouple the low-latency screen content communication problem into two parts, a scalable H.264 based encoding and an optimal scalable stream transmission scheduling. By leveraging the unique characteristics of screen content, a multi-layer scalable video encoding scheme is designed to achieve a certain error resilience while keeping good video coding efficiency. In the transmission scheduling module, an optimal frame skipping policy is proposed to schedule the frames in the buffer to maximize the play-out frame rate. In the performance evaluation, we simulate our system in both one-hop end-to-end topology and two-hop proxy-based topology. The simulation results show that the proposed scheme achieves much better performance on frame rate and average delay, especially in the low bandwidth condition.

In real-time screen remoting system, frame rate is one of essential factors that affect user experience. Therefore, how to compress diversity of screen contents fast and efficiently is a key issue. Existing video codecs such as H.264 are always used in such a system for screen compression. However, arbitrary-sized regions with large motion always exist in typical screen content videos, which lead to a lower encoding speed and higher bit-rate, thereby decrease the frame rate. This paper proposes an efficient motion detection algorithm, which is fast and efficient for large motion regions. In specific, a region-based motion detection is used to find motion vectors instead of traditional block based motion estimation. The motion vectors are then utilized by H.264 encoder for normal motion compensated prediction. Experimental results show that the proposed algorithm can reduce both encoding time and bit-rate significantly.

Modern digital family technology enables people surf the Internet and watch videos via a large screen. This paper proposes an effective scheme for using hand gestures rather than the common remote controllers to browse the web pages on a large TV screen. The proposed scheme models four gesture modes: mouse mode, scroll mode, zoom mode and input mode to help the user browse web pages naturally and comfortably. Then we combine RGB, depth, motion information and face detection to achieve accurate and real-time hand segmentation and gesture recognition for enabling the four gesture modes. The experiments show the proposed scheme works well in various illumination environments and complicated backgrounds with multiple moving humans. The recognition accuracy of hand shapes in the proposed scheme arrives at 98.50%, and the successful rate for visual digits input reaches 89.00%. Furthermore, the frame rate of the hand-gesture detection and recognition is about 18 fps. Thus the scheme is accurate, real-time and natural.

In this paper, we propose a layered screen video coding scheme based on existing video codecs to leverage hardware video codec for efficient screen video compression. In this scheme, the screen video compression is performed as two-layer coding: base layer coding and enhancement layer coding. The screen video is first analyzed in both frame and block levels for useful temporal and spatial information extraction to assist coding content selection in each layer. The non-skip screen frames are directly compressed by the conventional video codec in the base layer, while the screen contents sensitive to the video quality degradation are selected for improved coding in the enhancement layer. For contents to be enhanced, two intra coding modes are designed to improve the quality of the compressed text/graphics contents and suppress the artifacts introduced by chroma downsampling. The experimental results demonstrate that the screen video quality is improved objectively and subjectively by the proposed scheme with low cost on bitrate and computation complexity. Moreover, an average of 2.95dB coding gain is achieved in high bitrate.

Due to the divergent characteristics of image contents and text contents in screen videos, how to make the joint optimization leveraging rate-distortion (R-D) optimized block classification and bit allocation is critical to the compression performance. In this paper, a general model-based solution is proposed as an attempt to solve this problem. The contributions of this paper are twofold: First, the rate and distortion characteristics of image blocks and text blocks in block-based content-adaptive screen video encoder (BASC) are carefully studied, and the rate and distortion models are proposed. Second, with the proposed rate and distortion models, the R-D optimized block classification and bit allocation are derived using bisection searched Lagrange multiplier method. Experimental results demonstrate that the proposed R-D optimized block classification and bit allocation algorithms are able to adapt to diverse screen contents, which results in a significant gain of up to 4.5dB in PSNR.

Interactive screen sharing requires extremely low latency end-to-end transmission, which in turn requires highly efficient and low-complexity screen compression. In this paper, we present a block-based low-complexity screen compression scheme, in which multiple block modes are adopted to exploit the intra- and inter-frame redundancies. In particular, we classify the intra-coded blocks to pictorial blocks and textual blocks using a proposed fast block classification algorithm, which exploits the discriminative features between the pictorial and the textual blocks. Then, we design a low-complexity, yet efficient, algorithm to compress the textual blocks. We use base colors and escape colors to represent and quantize the textual pixels, which not only achieves high compression ratios but also preserves a high quality on textual pixels. The two-dimensionally predictive index coding and hierarchical pattern coding technologies are used to exploit local spatial correlations and global pattern correlation, respectively. To further utilize the correlation between the luminance and chrominance channels, we propose a joint-channel index coding method. We compare the coding efficiency and the computational complexity of the proposed scheme against the standard image coding schemes such as JPEG, JPEG2000, and PNG, the compound image compressor HJPC, and the popular video coding standard H.264. We also compare the visual quality of the proposed scheme against H.264 intra coding, JPEG2000, and HJPC. The evaluation results show that the proposed scheme achieves superior or comparable compression efficiency with much lower complexity than other schemes in most of the cases.

Unlike traditional RGB video, Kinect-like depth is characterized by its large variation range and instability. As a result, traditional video compression algorithms cannot be directly applied to Kinect-like depth compression with respect to coding efficiency. In this paper, we propose a lossy Kinect-like depth compression framework based on the existing codecs, aiming to enhance the coding efficiency while preserving the depth features for further applications. In the proposed framework, the Kinect-like depth is reformed first by divisive normalized bilateral filter (DNBL) to suppress the depth noises caused by disparity normalization, and then block-level depth padding is implemented for invalid depth region compensation in collaboration with mask coding to eliminate the sharp variation caused by depth measurement failures. Before the traditional video coding, the inter-frame correlation of reformed depth is explored by proposed 2D+T prediction, in which depth volume is developed to simulate 3D volume to generate pseudo 3D prediction reference for depth uniqueness detection. The unique depth region, called active region is fed into the video encoder for traditional intra and inter prediction with residual coding, while the inactive region is skipped during depth coding. The experimental results demonstrate that our compression scheme can save 55%-85% in terms of bit cost and reduce coding complexity by 20%-65% in comparison with the traditional video compression algorithms. The visual quality of the 3D reconstruction is also improved after employing our compression scheme.

Accuracy and stability of Kinect-like depth data is limited by its generating principle. In order to serve further applications with high quality depth, the preprocessing on depth data is essential. In this paper, we analyze the characteristics of the Kinect-like depth data by examing its generation principle and propose a spatial-temporal denoising algorithm taking into account its special properties. Both the intra-frame spatial correlation and the inter-frame temporal correlation are exploited to fill the depth hole and suppress the depth noise. Moreover, a divisive normalization approach is proposed to assist the noise filtering process. The 3D rendering results of the processed depth demonstrates that the lost depth is recovered in some hole regions and the noise is suppressed with depth features preserved.

The emergence of Kinect facilitates the possibility of depth capture in real-time and with low cost by consumers. It also provides powerful tool and inspiration for researchers to engage in new array of technology development. However, the quality of the depth map captured from Kinect is still inadequate for many applications due to holes, noises and artifacts existing within the depth information. In this paper, we present a texture assisted Kinect depth inpainting framework, aiming at obtaining improved depth information. In this framework, the relationship between texture and depth is investigated, and the characteristics of depth are also exploited. More specifically, texture edge information is extracted to assist the depth inpainting. Furthermore, filtering and diffusion are designed for hole-filling and edge alignment. Experiment results demonstrate that the Kinect depth can be appropriately repaired in both smooth and edge region. Comparing with the original depth, the inpainted depth information enhances the quality of advanced processing such as 3D reconstruction.

Screen is becoming a new dimension in cloud computing platforms, and low latency screen sharing in unreliable networks is becoming more and more important. Due to the different characteristics between the screen codecs and video codecs, current transmission technologies on low-latency video streaming cannot be directly applied to screen sharing. So in this paper we first theoretically analyze the difference in latency performance between ARQ and FEC for the UDP-based screen sharing. Then, considering the characteristics of the main-stream screen codecs, we propose an improved ARQ scheme to decrease the transmission latency. The experimental results show that the proposed system achieves better latency performance than the popular systems.

Compound video compression is crucial for remote control and data assessment. In this paper, we propose a content-aware layered video coding scheme as an attempt to efficiently compress the compound video. In this scheme, the compound video is analyzed and processed progressively at three pyramid levels: block, object and layer. Firstly, the compound video is analyzed by a block type classification technique to access each block's spatial and temporal properties. Secondly, the natural video object is detected adaptively in each frame based on the block type. Finally, the compound video content is distributed into different layers and specifically designed video coding algorithms are employed to compress each layer. Experiments demonstrate that our proposed scheme can preserve the advantages of the employed compression algorithms for each layer and outperform each of them in the compound video compression.

With the rapid development of depth data acquisition technology, the high precision depth becomes much easier to access in real-time by depth sensors, and the generated high dynamic range (HDR) depth is widely adopted to benefit the depth assistant applications. Accordingly, the HDR depth compression becomes essential for the efficient depth storage and transmission. In this paper, we introduce a layered compression framework for HDR depth to achieve efficient and low-complexity depth compression. To leverage the state-of-art 8-bit image/video encoders, the HDR depth is partitioned into two layers: most significant bit (MSB) layer and least significant bit (LSB) layer. For MSB layer, an error controllable pixel domain encoding scheme is proposed to guarantee the compatibility for existing 8-bit codec by controlling quantization errors added back to LSB layer. Meanwhile, the efficient major color extraction and adaptive quantization enhance the coding performance of MSB layer. For LSB layer, the layer data with limited dynamic range is compressed by normal 8-bit image/video based encoding scheme. The experimental results demonstrate that our coding scheme can achieve real-time depth compression with the satisfactory reconstruction quality. The encoding time is less than 31ms/frame and the decoding time is around 20ms/frame in average. Our compression scheme can be easily integrated into the real-time depth transmission system.

This paper presents a block-based low-complexity screen compression scheme. In this scheme, the input screen is split into non-overlapping blocks which are classified as pictorial blocks and textual blocks. We design a low-complexity yet efficient algorithm to compress the textual blocks. We use base colors plus escape pixels to represent and quantize the text pixels, and such quantization mechanism not only achieves high compression efficiency but also keeps low encoding/decoding complexity. We also propose the two-direction predictive index coding and hierarchical pattern coding technologies to utilize the local spatial correlation and the global pattern correlation for text pixels. In addition, to utilize the correlation between the luminance and chrominance channels, we propose a joint-channel index coding method to further improve the compression efficiency. The compression efficiency and complexity of the proposed method is evaluated against the popular image codecs JPEG/JPEG200 and PNG, a recent published screen compression scheme HJPC as well as the popular video codec H.264.

In this paper, we present a browser-friendly hybrid JPEG/PNG codec for compound images. First we employ a simple yet efficient block classification algorithm to identify the blocks to pictorial and textural ones. And then JPEG is used to encode the pictorial blocks and PNG is used for textual blocks. Our evaluation results show that our codec significantly outperforms JPEG and PNG in terms of rate-distortion performance, and also it outperforms JPEG, JPEG2000 and DjVu in terms of visual quality. Moreover, since JPEG and PNG are naturally supported by modern browsers, the coded images generated from our proposed coder can be natively supported by browsers and possible to be widely deployed in Web applications.

In this paper, we propose an efficient encoding solution for the screen-sharing applications with multiple clients connected. We first present a lightweight screen codec to compress the complicated screen content. Based on the architecture of the proposed codec, we propose a one-pass encoding algorithm for multiple bit-rates. The one-pass encoding algorithm enables the host to only involve in one-pass encoding process for multiple bitrates, and as a result the computation complexity of the host is decreased significantly. Specially, in the limited computing-resource case, the one-pass encoding algorithm can improve the screen-sharing performance by about 43%, in terms of framerates the clients can get. In addition, based on our codec, we propose a fast screen transcoding scheme for the data-center based screen-sharing applications.

In this paper, we present a proxy-based mobile web browser with rich experiences. We use the server-side web parsing and rendering to leverage the browser computing logic. We use a composite screen format to represent the display of the web content, incorporating the web background screen and the dynamic web objects. And then we employ a slice-based screen encoding scheme to efficiently compress the web background screen. Besides the display screen of the web content, we also send the side information of the web objects to enable the designed object-level interaction mechanisms. The experimental results show that our browser can achieve the superior browsing speed, compared with the native browser and yield much better visual quality than the existing proxy-based browser

In this paper, we present an interactive virtual display system to facilitate the ubiquitous user interaction with heterogeneous devices. By using small-size programmable hardware and wearable sensors, any display device (referred to as display surface) can act as a thin client for users to interact with the different remote devices. Under a flexible system architecture for local and remote devices' communication and collaboration, several techniques, such as adaptive screen compression, interactive ROI control, and accelerometer-based pointing input, are developed to improve the system performance and user experience. Evaluations show that the proposed system can efficiently utilize the remote computing resources and local display capabilities of ubiquitous devices, which will greatly benefit interactive multimedia applications.

In this paper, we propose a novel approach for high-dynamic-range (HDR) texture compression (TC) suitable for rendering systems of different capacities. Based on the previously proposed DHTC scheme, we first work out an improved joint-channel compression framework, which is robust and flexible enough to provide compressed HDR textures at different bit rates. Then, two compressed HDR texture formats based on the proposed framework are developed. The 8 bpp format is of near lossless visual quality, improving upon known state-of-the-art algorithms. And, to our knowledge, the 4 bpp format is the first workable 4 bpp solution with good quality. We also show that HDR textures in the proposed 4 bpp and 8 bpp formats can compose a layered architecture in the texture consumption pipeline, to significantly save the memory bandwidth and storage in real-time rendering. In addition, the 8 bpp format can also be used to handle traditional low dynamic range (LDR) RGBA textures. Our scheme exhibits a practical solution for compressing HDR textures at different rates and LDR textures with alpha maps.

In this paper, we propose a novel compression scheme for medical images with extended bit depth. Our scheme efficiently utilizes the VOI (value of interest) settings of the medical images, so that only a small part of the coded bit-stream is needed at the decoder to lossless display the original image with recommended VOI parameters. Gray-Golomb coding is used to support VOI progressive transmission. In addition, the pixel domain bit-plane coding enables accurate distortion estimation at pixel level for any truncated bit-stream. Compared to JPEG2000, our scheme achieves much higher VOI performance with very slight overall bit-rate increase.

In this paper, we propose a simple and effective scheme for screen image scaling, targeting at real-time applications like remote desktop and screen sharing. To balance visual quality and complexity, we build our scheme upon bilinear interpolation and devise a content adaptive post-processing method. Our scheme keeps the text/graphics regions of screen images as sharp as the original while avoiding magnifying noise and artifacts in the picture regions. Further, the involved operations are quite suitable for hardware acceleration. When implemented on commodity GPUs, our scheme achieves a frame rate more than 300 fps, which is fast enough for practical use.

In this paper, a joint complexity-distortion optimization approach is proposed for real-time H.264 video encoding under the power-constrained environment. The power consumption is first translated to the encoding computation costs measured by the number of scaled computation units consumed by basic operations. The solved problem is then specified to be the allocation and utilization of the computational resources. A computation allocation model (CAM) with virtual computation buffers is proposed to optimally allocate the computational resources to each video frame. In particular, the proposed CAM and the traditional hypothetical reference decoder model have the same temporal phase in operations. Further, to fully utilize the allocated computational resources, complexity-configurable motion estimation (CAME) and complexity-configurable mode decision (CAMD) algorithms are proposed for H.264 video encoding. In particular, the CAME is performed to select the path of motion search at the frame level, and the CAMD is performed to select the order of mode search at the macroblock level. Based on the hierarchical adjusting approach, the adaptive allocation of computational resources and the fine scalability of complexity control can be achieved.

In this paper, we propose a novel three-tiered network model for image hallucination based on the learnt knowledge composed of image patches relating low and high resolution. A common problem of previous hallucination methods is that irregularities are usually introduced into the constructed high-resolution images. We remove the irregularities in three steps. First, the hallucination with primal sketch priors is performed to construct a coarse high-frequency component. Second, enhancement is implemented to enforce local compatibility between the patches in the constructed component. Third, a Markov network is utilized to refine the enhanced high-frequency component. Experiments demonstrate that our model can hallucinate higher-quality images than existing methods.

This paper proposes a Wyner-Ziv (WZ) switching scheme for multiple bit-rate (MBR) video streaming over networks. Identical video content is encoded into a set of normal streams, which are generated by conventional hybrid video coding with multiple bit rates, so that streaming can dynamically switch among these normal streams according to available bandwidth. At encoder side, the WZ codec generates a switching stream by compressing the reconstructed frames of a certain normal stream that will be switched to, no matter which normal stream it switches from. At decoder side, for switching to the same frame, the same WZ switching stream is used to reconstruct the switching-to frame by taking the switching-from frame as the side information. The number of required WZ bits depends on the inherent mutual correlation between two frames switching to and from. Since the WZ switching streams are generated independently of the normal switching-from streams, given normal streams that can switch from any one to another, the proposed scheme reduces the number of switching streams from to . Furthermore, switching streams do not deteriorate the coding efficiency of normal streams when no switching occurs. However a big problem here, similar to requesting bits in distributed video coding, is how many WZ bits should be transmitted when a switching happens because the streaming scenario does not tolerate too much extra delay caused by the requests back and forth. Therefore, a Laplacian model, which is proved in the simplified case, is proposed to characterize the correlation between switching-to and switching-from frames. It can be used to accurately estimate the number of WZ bits at the server side.

Utilizing video correlations among views would definitely improve multiview video compression in terms of coding efficiency, which usually requests an expensive system to collect videos from different cameras and jointly compress them. Thanks to recent developments on distributed video coding, this paper proposes a new multiview video coding scheme based on Wyner-Ziv (WZ) coding technique, in which the complicated temporal and interview correlation exploration process is shifted from the encoder side to the decoder side so that broadband raw data traffic and high intensive computation for jointly encoding can be avoided. At the encoder side, a wavelet-based WZ scheme is proposed to compress video of every camera. Furthermore, in order to better utilize correlation in wavelet domain, all coefficients are organized as that done in SPIHT bit plane by bit plane. At the decoder side, a more flexible prediction technique that can jointly utilize temporal and view correlations is proposed to generate side information. Finally, experimental results show the proposed scheme significantly outperforms the conventional intra-frame coding for better random access and is even close to the inter-frame coding for better efficiency. Furthermore, compressed data is much robust when it is transmitted over an error-prone channel.

This paper proposes an efficient two-description image coding technique. The two side descriptions of an image are generated by quincunx subsampling. The decoding from any side description is done by an interpolation process that exploits sample correlation. Although the quincunx subsampling is a natural choice for the best use of sample correlations in image multiple-description coding (MDC), each side description is not amenable to existing image coding techniques because the pixels are not aligned rectilinearly. We show how this difficulty can be overcome by an adaptive directional lifting (ADL) transform that is particularly suitable for decorrelating samples on the quincunx lattice. The ADL transform can be embedded into JPEG 2000 to construct a practical MD image encoder. Experimental results demonstrate that the proposed image MDC scheme can achieve good coding performance.

This paper presents an efficient compound image compression approach based on H.264/AVC intra coding. The text blocks are distinguished from the picture blocks and compressed with a new coding mode. In particular, the text blocks are represented by base colors and index map in spatial domain. A color quantization algorithm optimized for compression is designed to generate this representation. As for the entropy coding of text blocks, a structure-aware context-based arithmetic coder is developed. The mode selection algorithm based on rate distortion optimization is used to select the text blocks along with H.264/AVC intra modes, which can adapt to the targeted bit-rate. Experimental results show that the proposed scheme can achieve 2.8dB gain on average for compound images compared with H.264/AVC intra coding.

A new pixel-domain distributed video coding (DVC) scheme is proposed in this paper, in which both the temporal and the spatial correlations are exploited only at the decoder. A video is treated as a collection of data correlated in temporal and spatial directions. Besides splitting a video into frames at different time instants, a frame is further split by spatially sub-sampling. Each yielded part is then encoded individually. At the decoder, the side information signals are from both adjacent frames and the spatially decoupled signals. To utilize these multiple side information signals, a new probability model is proposed, in which the transitional probability is calculated from the conditional probabilities on the multiple side information signals. The coding efficiency is enhanced by further removing the spatial redundancy, while the encoding complexity remains the same as the previous pixel-domain DVC techniques that only consider the temporal correlation

In this paper, we present a new image compression scheme, which is specially designed for computer generated compound color images. First we classify the image content into two kinds: text/graphic content and picture content. Then two different compression schemes are applied blocks of different types. We propose a two stage segmentation scheme which combines thresholding block features and rate-distortion optimization. The text/graphics blocks compression scheme consists of two parts: color quantization and lossless coding of quantized images. The input images will first be color quantized and converted to codebooks and labels, introducing constraint distortion to the color quantization images. Then generated labels and codebooks are lossless compressed respectively. We proposed a rate-distortion optimized color quantization algorithm for text/graphic content, which introduces distortion to text content and minimizes the bit rate produced by the following lossless entropy compression algorithm. The picture content is compressed using conventional image algorithms like JPEG. The results show that the proposed scheme achieves better coding performance than other images compression algorithms such as JPEG2000 and DjVu.

In this paper, we propose a joint power-distortion optimization scheme for real-time H.264 video encoding under the power constraint. Firstly, the power constraint is translated to the complexity constraint based on DVS technology. Secondly, a computation allocation model (CAM) with virtual buffers is proposed to facilitate the optimal allocation of constrained computational resource for each frame. Thirdly, the complexity adjustable encoder based on optimal motion estimation and mode decision is proposed to meet the allocated resource. The proposed scheme takes the advantage of some new features of H.264/AVC video coding tools such as early termination strategy in fast ME. Moreover, it can avoid suffering from the high overhead of the parametric power control algorithms and achieve fine complexity scalability in a wide range with stable rate-distortion performance. The proposed scheme also shows the potential of a further reduction of computation and power consumption in the decoding without any change on the existing decoders.

For a typical video distribution system, the video contents are first compressed and then stored in the local storage or transmitted to the end users through networks. While the compressed videos are transmitted through error-prone networks, error robustness becomes an important issue. In the past years, a number of rate-distortion (R-D) optimized coding mode selection schemes have been proposed for error-resilient video coding, including a recursive optimal per-pixel estimate (ROPE) method. However, the ROPE-related approaches assume integer-pixel motion-compensated prediction rather than subpixel prediction, whose extension to H.264 is not straightforward. Alternatively, an error-robust R-D optimization (ER-RDO) method has been included in H.264 test model, in which the estimate of pixel distortion is derived by simulating decoding process multiple times in the encoder. Obviously, the computing complexity is very high. To address this problem, we propose a new end-to-end distortion model for R-D optimized coding mode selection, in which the overall distortion is taken as the sum of several separable distortion items. Thus, it can suppress the approximation errors caused by pixel averaging operations such as subpixel prediction. Based on the proposed end-to-end distortion model, a new Lagrange multiplier is derived for R-D optimized coding mode selection in packet-loss environment by taking into account of the network conditions. The rate control and complexity issues are also discussed in this paper

Energy supply is always a bottleneck in wireless video applications. To prolong the battery life, a typical wireless video encoding and streaming system should have the capability of dynamically adjusting its encoding parameters based on the remaining battery supply. In this paper, we consider two types of application scenarios as well as the associated optimization problems. The first one is, given a video quality level, how to find out the optimized encoding parameters so that the total power consumption is minimized. The second one is, given a fixed power consumption level, how to calculate the optimized encoding parameters so that the video quality is best. Accordingly, we jointly consider the power consumption and video quality and pose the above two problems into a uniform optimization framework. Under this framework, we give an efficient solution for the MPEG-4 AVC/H.264 codec and Wi-Fi transmission devices based on a detailed complexity analysis of the H.264 encoder. The analysis method can be easily extended to other video encoder and transmission configuration.

In this paper, we propose a Wyner-Ziv video coding scheme based on set-partitioning in hierarchical trees (SPIHT) which can utilizing not only the spatial and temporal correlations but also the high-order statistical correlations. Wyner-Ziv theory on source coding with side information is employed as the basic coding principle, which makes the independent encoding and joint decoding become possible. In the proposed scheme, wavelet transform is first used to de-correlate the spatial dependency of a Wyner-Ziv frame. Then the quantized transform coefficients are organized by using magnitude with a set partitioning sorting algorithm. The ordered bit planes are coded using the Wyner-Ziv coding based on turbo codes. At the decoder, side information generated by motion compensated interpolation is used to conditionally decode the Wyner-Ziv frame. Since the high order statistical correlation is used, the proposed algorithm owns advantages over the traditional pixel-domain and transform-domain Wyner-ziv video coding schemes.

In this paper, we propose a novel bit-stream switching scheme for the multiple bit-rate (MBR) video streaming, in which a Wyner-Ziv coded frame is used to overcome the mismatch between the MBR streams when the switching occurs. With the proposed technique, the MBR streams can be independently encoded without losing any coding efficiency. Similar to distributed video coding, the proposed Wyner-Ziv switching scheme also faces the challenge of rate allocation at the server side. To solve this problem, we propose a new correlation model based on the analysis on the reconstructed frames from the streams with different bit rates. Accordingly, the number of transmitted bits can be on-line calculated based on the correlation model without any feedback from the decoder. With the proposed technique, the actually transmitted Wyner-Ziv bits are only few more than the truly requested bits. However, the delay due to the bit requesting process can be avoided.

It has been commonly recognized that multiple bit-rate (MBR) encoding provides a concise method for video streaming over bandwidth-fluctuant networks. The key problem of the MBR technique lies in how to seamlessly switch one bit-stream to another one. To tackle this problem, we propose a bit-stream switching framework based on the Wyner-Ziv coding. Within the propose framework, the multiple bit-streams can be individually encoded without data exchange, which also supports the random switching at any desired frame without affecting the original coding efficiency of the regular bit-stream. In particular, two different implementation schemes under the same framework are presented. Different from the traditional switching schemes, the proposed method can use the same switching frame for the switching from any other bit-stream to the current one, which means less storage and less encoding efforts. Simulation results and comparison between the proposed method and the traditional switching method in H.264 are also presented

This paper proposes a distributed video coding scheme based on the zero tree entropy (ZTE) coding. Wyner-Ziv theory on source coding with side information is taken as the basic coding principle, which makes independent encoding and joint decoding possible. In this scheme, wavelet transform is used to exploit the spatial correlation of a Wyner-Ziv frame. The quantized wavelet coefficients are reorganized in terms of the zero tree structure so as to identify the significant and insignificant coefficients. The significance map is intra-codec and transmitted. In particular, the significant coefficients are independently encoded with turbo coder, and only the parity bits are transmitted. At the decoder, a predictive frame generated through motion-compensated prediction is used as the side information, with which the Wyner-Ziv frame can be conditionally decoded. Experimental results show that, compared to the traditional intra-frame coding and pixel-domain Wnyer-Ziv video coding, the proposed scheme can achieve a better coding performance, especially at low bit rates.

Global disparity between views is usually caused by the displacement between cameras, which can be accurately represented by a global geometric transformation. In this paper, we first propose an inter-view motion model in terms of the global geometric transformation to represent the motion correlation between two adjacent views. Specifically, the motion vector of a pixel in one view may be directly derived from that in another view according to the inter-view motion model. Further, we propose an inter-view direct mode to signal the decoder that the motion of a macroblock (MB) can be achieved from the coded view without any coding bits. The proposed inter-view direct mode is further incorporated in the existing multiview video coding (MVC) schemes (i.e., AVC-based MVC and 4-D wavelet-based MVC), working together with the other classical coding modes. The mode selection at each MB is accomplished with the rate-distortion optimization technique. The proposed inter-view direct mode can significantly reduce bits to code motion vectors especially at low bit rates, thus improving the coding efficiency

The conventional multiview video coding (MVC) schemes, utilizing both neighboring temporal frames and view frames as possible references, have only shown a slight gain over those using temporal frames alone in terms of coding efficiency. The reason for this is that the neighboring temporal frames exhibit stronger correlation with the current frame and the view frames often fail to be selected as references. This paper proposes an elegant MVC framework using high dimensional wavelet, which rightly matches the inherent high dimension property of multiview video. It also makes a better usage of both temporal and view correlations thanks to the hierarchical decomposition. Besides the proposed framework, this paper also investigates MVC coding from the following aspects. First, a disparity-compensated view filter (DCVF) with pixel alignment is proposed, which can accommodate both global and local view disparities among view frames. The proposed DCVF and the existing motion-compensated temporal filter (MCTF) unify the view and temporal decompositions as a generic lifting transform. Second, an adaptive decomposition structure based on the analysis of the temporal and view correlations is proposed. A Lagrangian cost function is derived to determine the optimum decomposition structure. Third, the major components of the proposed MVC coding are figured out, including macroblock type design, subband coefficient coding, and rate allocation. Extensive experiments are carried out on the MPEG 3DAV test sequences and the superior performance of the proposed MVC coding is demonstrated. In addition, the proposed MVC framework can easily support temporal, spatial, SNR, as well as view scalabilities

In this paper, an efficient rate-control scheme for H.264/AVC video encoding is proposed. The redesign of the quantization scheme in H.264/AVC results in that the relationship between the quantization parameter and the true quantization stepsize is no longer linear. Based on this observation, we propose a new rate-distortion (R-D) model by utilizing the true quantization stepsize and then develop an improved rate-control scheme for the H.264/AVC encoder based on this new R-D model. In general, the current R-D optimization (RDO) mode-selection scheme in H.264/AVC test model is difficult for rate control, because rate control usually requires a predetermined set of motion vectors and coding modes to select the quantization parameter, whereas the RDO does in the different order and requires a predetermined quantization parameter to select motion vectors and coding modes. To tackle this problem, we develop a complexity-adjustable rate-control scheme based on the proposed R-D model. Briefly, the proposed scheme is a one-pass process at frame level and a partial two-pass process at macroblock level. Since the number of macroblocks with the two-pass processing can be controlled by an encoder parameter, the fully one-pass implementation is a subset of the proposed algorithm. An additional topic discussed in this paper is about video buffering. Since a hypothetical reference decoder (HRD) has been defined in H.264/AVC to guarantee that the buffers never overflow or underflow, the more accurate rate-allocation schemes are proposed to satisfy these requirements of HRD.

To leave the maximum flexibility in encoder to optimize the trade-off between coding performance and complexity, in video coding standards such as H.264 AVC, H.263 and MPEG-4 etc, any number of B pictures and any arrangement of P pictures within a group of pictures (GOP) of arbitrary length are permitted. In addition, multiple reference picture prediction is also permitted in some video coding systems such as H.264/AVC to achieve efficient coding by allowing the encoder to select reference pictures among a large number of coded pictures. Both of the above cases without fixing the temporal distance between forward and backward reference pictures require the division operation for deriving the motion vectors of direct mode, which can efficiently exploit the temporal correlation among pictures and does not require any bits for coding the motion vectors. However, the division is an expensive and undesired operation in video decoder hardware design. Although the H.264/AVC video standard has provided a scaling technique to tackle this problem, unfortunately, its performance is also deteriorated. This paper presents a new scaling technique to both remove the division operation for deriving direct mode motion vectors and efficiently improve the accuracy of derived direct mode motion vectors compared with the scaling technique in H.264/AVC.

In this paper, an efficient rate-distortion (RD) model for an H.264 video encoder in a packet loss environment is presented. The encoder keeps tracking the potential error propagation on a block basis by taking into account the source characteristics, network conditions as well as the error concealment method. The end-to-end distortion invoked in this RD model is estimated according to the potential error-propagated distortion stored in a distortion map. The distortion map, in terms of each frame, is derived after the frame is encoded, which can be used for the RD-based encoding of the subsequent frames. Since the channel distortion has been considered in the proposed RD model, the new Lagarangian parameter is derived accordingly. The proposed method outperforms the error robust rate-distortion optimization method in the H.264 test model better in terms of both transmission efficiency and computational complexity.

In the conventional video codec, B pictures usually employ five prediction modes, including bi-prediction, direct prediction, forward prediction, backward prediction and intra modes. Among them, the direct and bi-prediction modes, whose prediction values are achieved from both forward and backward reference pictures, are more efficient in exploiting the temporal correlation between the reference pictures and the current B picture. In addition, the direct mode does not require any bits for coding the motion vectors. Therefore, the blocks coded with direct or bi-predictive mode are usually much more than the other modes. To further take advantage of the bi-prediction, We propose a new bi-predictive coding technique, which can achieve a good tradeoff between the bit-rate saving for motion vector coding and the prediction accuracy. Moreover, we also propose spatial motion vector prediction and motion vector scaling techniques to improve the accuracy of the derived direct mode motion vectors. All these techniques have been adopted in the AVS video coding standard.

We propose several improved error concealment (EC) algorithms based on the H.264/AVC non-normative decoder. The major differences are that motion compensated EC is introduced for intra frames, whereas spatial EC is introduced for inter frames. As for the EC of intra frames, scene change detection, motion activity detection and MV retrieval are hierarchically performed to decide whether spatial or temporal information is to be used. As for the EC of inter frames, scene change is also detected to avoid merging the scenes from different video shots. Therefore, the main idea of the proposed algorithms is that both spatial and temporal correlations are utilized for the EC of intra and inter frames. Both subjective and objective simulations under Internet conditions show that the proposed algorithm greatly outperforms that in the H.264/AVC non-normative decoder.

It is well observed that usually the run-length of successive zero coefficients becomes longer and the magnitude of non-zero coefficients gets smaller while the DCT subband frequency increases. Therefore, the probability distributions of level/run combinations should be different at different DCT positions. Based on this observation, an efficient context based 2D-VLC entropy coder, especially designed for 8/spl times/8 DCT, is proposed to exploit this context characteristic. The key element of the proposed coder is to design multiple 2D-VLC tables and each can be used context-adaptively to match different level/run combination probability distributions better. Another key element is the usage of exponential-Golomb codes, which keep the multiple tables with low memory requirement. In terms of coding efficiency, the experimental results show that the proposed method can gain up to 0.23 dB when compared to the one-table-for-one-block coding method.

Intra-prediction is a widely used technique in intra coding. In H.264, nine directional prediction modes are used, and in AVS (audio-video coding standard of China), six modes are used. The paper proposes a multiple mode, intra-prediction scheme to achieve a better balance between prediction precision and side information cost. Three kinds of additional modes are introduced to increase prediction precision from different aspects. A significant performance gain can be achieved using this multiple mode intra-prediction, with little complexity increase.

The JVT video coding scheme (MPEG-4 AVC/H.264) is a promising technique due to its high coding efficiency. Hypothetical reference decoder (HRD) is a very important part in JVT video coding, which represents a set of normative requirements on bitstream for the purpose of avoiding buffer overflow and underflow. The problem of HRD requirements can be solved by rate control. This paper proposes an effective rate control scheme for JVT video coding with HRD considerations. First, bit allocation with HRD constraints is presented, and second, based on a simple rate distortion model, a single pass rate control is implemented on both frame level and macroblock level. Experimental results show that the proposed rate control algorithm can achieve the target bit rate with very little bit rate or image quality fluctuation, and meanwhile it can well meet the HRD requirements. Furthermore, the proposed algorithm is so simple that it only introduces little computation complexity. Therefore, it can be used in real time video coding.

Spatial prediction is a promising technique for image coding. For example, the coming AVC/H.264 standard adopts directional spatial prediction in the intra frame coding. Similar to the traditional DCT-based image coding schemes, it still scan the transform coefficients in a zigzag order, which is inefficient for coding the residual signals predicted from the different directions. To tackle this problem, a new scheme of scanning the transform coefficients by utilizing the spatial prediction information is proposed in this paper. The distribution of transform coefficients from each direction of spatial predictions is fully studied. According to the statistics, the adaptive scan table is derived for each type of spatial predictions, which is indicated by the prediction mode. Experimental results demonstrate that the proposed scheme can always outperform the JVT codec using zigzag scanning. Moreover, it does not introduce any extra computing costs in software implementation.

Hypothetical reference decoder (HRD) is a mathematical model of a decoder and its input buffer. It represents a set of normative requirements on bitstream for the purpose of avoiding buffer overflow and underflow. In other words, any coded video bitstream shall meet constraints imposed by HRD model. Two HRD models, i.e. the earliest arrival time leaky bucket (EAT- LB) and the constrained arrival time leaky bucket (CAT-LB), have been proposed for the JVT standard jointly developed by ISO/IEC and ITU-T. EAT-LB has the lower initial delay, whereas the CAT-LB model has the lower maximum delay. This paper proposes an improved leaky bucket model called latest arrival time leaky bucket (LAT-LB) so as to achieve the advantages of EAT-LB and CAT-LB simultaneously. The proposed LAT-LB first defines the data transmission schedule according to a set of resume points and stop points, and then derives the HRD parameters according to this schedule. Experimental results demonstrate that the proposed LAT-LB outperforms EAT-LB and CAT-LB in terms of initial delay and maximum delay.

An efficient static sprite-generation and compression scheme is proposed for background video coding. First, considering both short- and long-term motion influences, a hybrid technique is proposed in the sprite generation to search the background motion relative to the generated sprite image. In order to prevent the sprite image from being blurred due to the inaccurate segmentation, a reliability-based blending technique is developed. Both the background video at every frame and the sprite image are categorized into reliable, unreliable, and undefined regions. Each region uses a unique strategy in blending. This makes the proposed technique work well even in the case of coarse segmentation. Second, an arbitrary-shape spatial prediction technique is proposed to efficiently compress the sprite image. Due to the property of the background video object, the sprite image often contains many flat and texture-correlation regions. Therefore, utilizing directional spatial prediction can significantly improve the efficiency of sprite coding. In general, the generated sprite is not a rectangular image. Since traditional spatial-prediction methods have difficulty in handling those contour blocks, a padding technique derived from rate-distortion optimization is proposed. Experimental results show that the proposed scheme can generate the background sprite image with better visual quality. Compared with the MPEG-4 sprite-coding scheme, it can also significantly improve the coding efficiency up to 3.0 dB at low bit rates.

Sprite is an efficient and concise method for the representation of a background video object, which is typically compressed with the MPEG-4 object-based coding technique. Due to the property of a background video object, a sprite image often comprises many flat and texture-correlation regions. In order to fully exploit spatial redundancy, the paper proposes an idea of utilizing directional spatial prediction to improve sprite coding. In general, the generated sprite is not a rectangular image. Since traditional spatial prediction techniques have difficultly dealing with those contour blocks, a padding technique is first proposed to fill transparent regions in contour blocks by considering both the correlation of pixels within a block and the correlation among neighboring blocks. Then every block is encoded by the INTRA coding technique developed in H.26L. Experimental results show that the propose sprite coding scheme outperforms the MPEG-4 object-based coding by up to 3.0 dB at low bit rates. Furthermore, the proposed scheme essentially extends directional spatial prediction from frame-based coding to arbitrary shape object coding.

Sprite coding is a very efficient method for the representation and compression of background video object, which has been accepted by MPEG-4 standard. However, how to automatically generate sprite image from raw video sequence is still an open issue, because the motion of foreground video objects not only disturbs the accuracy of motion estimation but also blurs the generated sprite image. This paper presents a novel technique for the generation of background sprite image with improved global motion estimation (GME) and automatic segmentation. Since GME plays an important role in sprite generation, a hybrid scheme jointing the short-term and long-term motion estimation techniques is first developed in this paper. Automatic video segmentation based on available motion information is incorporated into sprite generation, which can mask moving foreground objects and meanwhile protect background sprite from blur. Therefore, the proposed technique is used to construct high quality sprite directly from a raw video sequence. Furthermore, with this technique the existing sprite coding can be extended to the field of frame-based coding.