The majority of standard diffusion models employ pixel-wise degradations while neglecting multi-scale characteristics of images. Recently, generalized diffusion models with Positive Semi-definite Degradations (PSD), such as heat dissipation and blurring, have been proposed to solve it, but suffering from problems of low generation quality due to incomplete optimization analysis and non-adaptiveness to the training process and different data distributions with hand-crafted and fixed inductive biases. In this paper, we present a comprehensive theoretical analysis of the optimization process in frequency domain for PSD-based generalized diffusion models, which implies the forward process of PSD frequency domain non-isotropic degradation implicitly acting on the inductive biases of the Variational Lower Bound non-isotropic weighting in the optimization reverse process. Based on this insight, we propose the Frequency Inductive Biases Bootstrapping Optimization (FIBBO) method, which parameterizes the forward process and learns distinct frequency degradation-generation trajectories iteratively. To tackle the problem of PSD hand-crafted and fixed inductive biases, FIBBO dynamically modifies the non-isotropic Gaussian kernel of the forward degradation process so that the inductive biases introduced can be adjusted adaptively during training. Experiments on public datasets show that FIBBO makes significant improvements in the generation quality of PSD-based generalized diffusion models.

Spike cameras generate binary spikes in response to light intensity changes, enabling high-speed visual perception with unprecedented temporal resolution. However, the unique characteristics of spike stream present significant challenges for reconstructing dense 3D scene representations, particularly in dynamic environments and under non-ideal lighting conditions. In this paper, we introduce DSNeRF, the first method to derive a NeRF-based volumetric scene representation from spike camera data. Our approach leverages NeRF's multi-view consistency to establish robust self-supervision, effectively eliminating erroneous measurements and uncovering coherent structures within exceedingly noisy input amidst diverse real-world illumination scenarios. We propose a novel mapping from pixel rays to the spike domain, integrating the spike generation process directly into NeRF training. Specifically, DSNeRF introduces an integrate-and-fire neuron layer that models non-idealities to capture intrinsic camera noise, including both random and fixed-pattern spike noise, thereby enhancing scene fidelity. Additionally, we propose a motion-guided spiking neuron layer and a long-term rendering photometric loss to better align dynamic spike streams, ensuring accurate scene geometry. Our method optimizes neural radiance fields to render photorealistic novel views from continuous spike streams, demonstrating advantages over other vision sensors in certain scenes. Empirical evaluations on both real and simulated sequences validate the effectiveness of our approach.

Recent advances in surface reconstruction for 3D Gaussian Splatting (3DGS) have enabled remarkable geometric accuracy. However, their performance degrades in photometrically ambiguous regions such as reflective and textureless surfaces, where unreliable cues disrupt photometric consistency and hinder accurate geometry estimation. Reflected light is often partially polarized in a manner that reveals surface orientation, making polarization an optical complement to photometric cues in resolving such ambiguities. Therefore, we propose PolarGS, an optics-aware extension of RGB-based 3DGS that leverages polarization as an optical prior to resolve photometric ambiguities and enhance reconstruction accuracy. Specifically, we introduce two complementary modules: a polarization-guided photometric correction strategy, which ensures photometric consistency by identifying reflective regions via the Degree of Linear Polarization (DoLP) and refining reflective Gaussians with Color Refinement Maps; and a polarization-enhanced Gaussian densification mechanism for textureless area geometry recovery, which integrates both Angle and Degree of Linear Polarization (A&DoLP) into a PatchMatch-based depth completion process. This enables the back-projection and fusion of new Gaussians, leading to a more complete reconstruction. PolarGS is framework-agnostic and achieves superior geometric accuracy compared to state-of-the-art methods.

Generating 3D-based body movements from speech shows great potential in extensive downstream applications, while it still suffers challenges in imitating realistic human movements. Predominant research efforts focus on end-to-end generation schemes to generate co-speech gestures, spanning GANs, VQ-VAE, and recent diffusion models. As an ill-posed problem, in this paper, we argue that these prevailing learning schemes fail to model crucial inter- and intra-correlations across different motion units, i.e. head, body, and hands, thus leading to unnatural movements and poor coordination. To delve into these intrinsic correlations, we propose a unified Hierarchical Implicit Periodicity (HIP) learning approach for audio-inspired 3D gesture generation. Different from predominant research, our approach models this multi-modal implicit relationship by two explicit technique insights: i) To disentangle the complicated gesture movements, we first explore the gesture motion phase manifolds with periodic autoencoders to imitate human natures from realistic distributions while incorporating non-period ones from current latent states for instance-level diversities. ii) To model the hierarchical relationship of face motions, body gestures, and hand movements, driving the animation with cascaded guidance during learning. We exhibit our proposed approach on 3D avatars and extensive experiments show our method outperforms the state-of-the-art co-speech gesture generation methods by both quantitative and qualitative evaluations. Code and models will be publicly available.

Motion quantization codebooks have been widely adopted to facilitate co-speech motion generation. However, the conventional quantization-based generation paradigm-which relies on probabilistic token sampling from limited discrete codebooks-suffers from two major limitations: crude, unreasonable motion representations and fixed, homogenized motion token sequences. To overcome these issues, we propose a novel explicit generation paradigm based on generative continuous quantization. Specifically, we first introduce a continuous quantization method to derive a set of generative motion units. This approach enables smoother and more accurate representation of human motion compared to classical methods. Building on these generative units, we further propose a compositional weight generation paradigm that replaces probabilistic sampling with deterministic, explicit motion synthesis. Moreover, as generalization capability is crucial for real-world deployment, we design a fully audio-aware encoder to extract style features that are decoupled from content. These features are integrated into the motion decoder via Adaptive Instance Normalization to enhance cross-speaker facial style generalization. Our method achieves state-of-the-art performance on two public datasets. Notably, owing to its concise and efficient architecture, our model attains an inference speed exceeding 4000 fps on the SHOW dataset, demonstrating strong potential for practical real-time applications.

Neural Radiance Fields can achieve photo-realistic rendering results, but the occlusion in front of the target object is a common and extreme scenario in practice that cannot be neglected. The prevailing works attempt to remove the occlusions using external 2D visual priors, which are not constrained to provide 3D-consistent guidance for the specific scenarios. In this paper, we propose UncNeRF, which utilizes multi-view clues from captured defective images to uncover the heavily occluded object. Specifically, we provide additional multi-view complementary optimization supervisions using object-centric forward warping and enhance the target object reconstruction by sampling pseudo-training views and introducing external spatial-relation regularization. To evaluate the reconstruction performance of occluded objects, we present the challenging and diverse Heavy Occlusion Removal (HOR) dataset consisting of synthetic and real-world scenes, whose target objects to be reconstructed are heavily occluded. Experimental results show that our method achieves state-of-the-art performance in heavy occlusion removal compared to other methods.

In-context learning (ICL), a predominant trend in instruction learning, aims at enhancing the performance of large language models by providing clear task guidance and examples, improving their capability in task understanding and execution. This paper investigates ICL on Large Vision-Language Models (LVLMs) and explores the policies of multi-modal demonstration selection. Existing research efforts in ICL face significant challenges: First, they rely on pre-defined demonstrations or heuristic selecting strategies based on human intuition, which are usually inadequate for covering diverse task requirements, leading to sub-optimal solutions; Second, individually selecting each demonstration fails in modeling the interactions between them, resulting in information redundancy. Unlike these prevailing efforts, we propose a new exploration-exploitation reinforcement learning framework, which explores policies to fuse multi-modal information and adaptively select adequate demonstrations as an integrated whole. The framework allows LVLMs to optimize themselves by continually refining their demonstrations through self-exploration, enabling the ability to autonomously identify and generate the most effective selection policies for in-context learning. Experimental results verify the superior performance of our approach on four Visual Question-Answering (VQA) datasets, demonstrating its effectiveness in enhancing the generalization capability of few-shot LVLMs.

Class-Incremental Semantic Segmentation (CISS) requires continuous learning of newly introduced classes while retaining knowledge of past classes. By abstracting mainstream methods into two stages (visual feature extraction and prototype-feature matching), we identify a more fundamental challenge termed catastrophic semantic entanglement. This phenomenon involves Prototype-Feature Entanglement caused by semantic misalignment during the incremental process, and Background-Increment Entanglement due to dynamic data evolution. Existing techniques, which rely on visual feature learning without sufficient cues to distinguish targets, introduce significant noise and errors. To address these issues, we introduce a Language-inspired Bootstrapped Disentanglement framework (LBD). We leverage the prior class semantics of pre-trained visual-language models (e.g., CLIP) to guide the model in autonomously disentangling features through Language-guided Prototypical Disentanglement and Manifold Mutual Background Disentanglement. The former guides the disentangling of new prototypes by treating hand-crafted text features as topological templates, while the latter employs multiple learnable prototypes and mask-pooling-based supervision for background-incremental class disentanglement. By incorporating soft prompt tuning and encoder adaptation modifications, we further bridge the capability gap of CLIP between dense and sparse tasks, achieving state-of-the-art performance on both Pascal VOC and ADE20k, particularly in multi-step scenarios.

Anomaly detection is essential for the safety and reliability of autonomous driving systems. Current methods often focus on detection accuracy but neglect response time, which is critical in time-sensitive driving scenarios. In this paper, we introduce real-time anomaly detection for autonomous driving, prioritizing both minimal response time and high accuracy. We propose a novel multimodal asynchronous hybrid network that combines event streams from event cameras with image data from RGB cameras. Our network utilizes the high temporal resolution of event cameras through an asynchronous Graph Neural Network and integrates it with spatial features extracted by a CNN from RGB images. This combination effectively captures both the temporal dynamics and spatial details of the driving environment, enabling swift and precise anomaly detection. Extensive experiments on benchmark datasets show that our approach outperforms existing methods in both accuracy and response time, achieving millisecond-level real-time performance.

Semantic segmentation has achieved great success in ideal conditions. However, when facing extreme conditions (e.g., insufficient light, fierce camera motion), most existing methods suffer from significant information loss of RGB, severely damaging segmentation results. Several researches exploit the high-speed and high-dynamic event modality as a complement, but event and RGB are naturally heterogeneous, which leads to feature-level mismatch and inferior optimization of existing multi-modality methods. Different from these researches, we delve into the edge secret of both modalities for resilient fusion and propose a novel Edge-awareness Semantic Concordance framework to unify the multi-modality heterogeneous features with latent edge cues. In this framework, we first propose Edge-awareness Latent Re-coding, which obtains uncertainty indicators while realigning event-RGB features into unified semantic space guided by re-coded distribution, and transfers event-RGB distributions into re-coded features by utilizing a pre-established edge dictionary as clues. We then propose Re-coded Consolidation and Uncertainty Optimization, which utilize re-coded edge features and uncertainty indicators to solve the heterogeneous event-RGB fusion issues under extreme conditions. We establish two synthetic and one real-world event-RGB semantic segmentation datasets for extreme scenario comparisons. Experimental results show that our method outperforms the state-of-the-art by a 2.55% mIoU on our proposed DERS-XS, and possesses superior resilience under spatial occlusion. Our code and datasets are publicly available at https://github.com/iCVTEAM/ESC.

Recent advances in 3D Gaussian Splatting (3DGS) have improved the visual fidelity of dynamic avatar reconstruction. However, existing methods often overlook the inherent chromatic similarity of human skin tones, leading to poor capture of intricate facial geometry under subtle appearance changes. This is caused by the affine approximation of Gaussian projection, which fails to be perspective-aware to depth-induced shear effects. To this end, we propose True-to-Geometry Avatar Dynamic Reconstruction (TGA), a perspective-aware 4D Gaussian avatar framework that sensitively captures fine-grained facial variations for accurate 3D geometry reconstruction. Specifically, to enable color-sensitive and geometry-consistent Gaussian representations under dynamic conditions, we introduce Perspective-Aware Gaussian Transformation that jointly models temporal deformations and spatial projection by integrating Jacobian-guided adaptive deformation into the homogeneous formulation. Furthermore, we develop Incremental BVH Tree Pivoting to enable fast frame-by-frame mesh extraction for 4D Gaussian representations. A dynamic Gaussian Bounding Volume Hierarchy (BVH) tree is used to model the topological relationships among points, where active ones are filtered out by BVH pivoting and subsequently re-triangulated for surface reconstruction. Extensive experiments demonstrate that TGA achieves superior geometric accuracy.

Few-Shot Class-Incremental Learning (FSCIL) challenges models to sequentially learn new classes from minimal examples without forgetting prior knowledge, a task complicated by the stability-plasticity dilemma and data scarcity. Current FSCIL methods often struggle with generalization due to their reliance on limited datasets. While diffusion models offer a path for data augmentation, their direct application can lead to semantic misalignment or ineffective guidance. This paper introduces Diffusion-Classifier Synergy (DCS), a novel framework that establishes a mutual boosting loop between diffusion model and FSCIL classifier. DCS utilizes a reward-aligned learning strategy, where a dynamic, multi-faceted reward function derived from the classifier's state directs the diffusion model. This reward system operates at two levels: the feature level ensures semantic coherence and diversity using prototype-anchored maximum mean discrepancy and dimension-wise variance matching, while the logits level promotes exploratory image generation and enhances inter-class discriminability through confidence recalibration and cross-session confusion-aware mechanisms. This co-evolutionary process, where generated images refine the classifier and an improved classifier state yields better reward signals, demonstrably achieves state-of-the-art performance on FSCIL benchmarks, significantly enhancing both knowledge retention and new class learning.

Depicting novel classes with language descriptions by observing few-shot samples is inherent in human-learning systems. This lifelong learning capability helps to distinguish new knowledge from old ones through the increase of open-world learning, namely Few-Shot Class-Incremental Learning (FSCIL). Existing works to solve this problem mainly rely on the careful tuning of visual encoders, which shows an evident trade-off between the base knowledge and incremental ones. Motivated by human learning systems, we propose a new Language-inspired Relation Transfer (LRT) paradigm to understand objects by joint visual clues and text depictions, composed of two major steps. We first transfer the pretrained text knowledge to the visual domains by proposing a graph relation transformation module and then fuse the visual and language embedding by a text-vision prototypical fusion module. Second, to mitigate the domain gap caused by visual finetuning, we propose context prompt learning for fast domain alignment and imagined contrastive learning to alleviate the insufficient text data during alignment. With collaborative learning of domain alignments and text-image transfer, our proposed LRT outperforms the state-of-the-art models by over 13% and 7% on the final session of miniImageNet and CIFAR-100 FSCIL benchmarks.

Spike cameras, leveraging spike-based integration sampling and high temporal resolution, offer distinct advan-tages over standard cameras. However, existing approaches reliant on spike cameras often assume optimal illumination, a condition frequently unmet in real-world scenarios. To address this, we introduce SpikeNeRF, the first work that derives a NeRF-based volumetric scene representation from spike camera data. Our approach leverages NeRF's multi-view consistency to establish robust self-supervision, effectively eliminating erroneous measurements and uncovering coherent structures within exceedingly noisy input amidst diverse real-world illumination scenarios. The framework comprises two core elements: a spike generation model incorporating an integrate-and-fire neuron layer and parameters accounting for non-idealities, such as threshold variation, and a spike rendering loss capable of general-izing across varying illumination conditions. We describe how to effectively optimize neural radiance fields to render photorealistic novel views from the novel continuous spike stream, demonstrating advantages over other vision sen-sors in certain scenes. Empirical evaluations conducted on both real and novel realistically simulated sequences affirm the efficacy of our methodology. The dataset and source code are released at https://github.com/BIT-Vision/SpikeNeRF.

Neural Radiance Fields (NeRF) achieves impressive 3D representation learning and novel view synthesis results with high-quality multi-view images as input. However, motion blur in images often occurs in low-light and high-speed motion scenes, which significantly degrades the reconstruction quality of NeRF. Previous deblurring NeRF methods struggle to estimate pose and lighting changes during the exposure time, making them unable to accurately model the motion blur. The bio-inspired event camera measuring intensity changes with high temporal resolution makes up this information deficiency. In this paper, we propose Event-driven Bundle Adjustment for Deblurring Neural Radiance Fields (EBAD-NeRF) to jointly optimize the learnable poses and NeRF parameters by leveraging the hybrid event-RGB data. An intensity-change-metric event loss and a photo-metric blur loss are introduced to strengthen the explicit modeling of camera motion blur. Experiments on both synthetic and real-captured data demonstrate that EBAD-NeRF can obtain accurate camera trajectory during the exposure time and learn a sharper 3D representations compared to prior works.

Accurate environment dynamics modeling is crucial for obtaining effective state representations in visual reinforcement learning (RL) applications. However, when facing multiple input modalities, existing dynamics modeling methods (e.g., DeepMDP) usually stumble in addressing the complex and volatile relationship between different modalities. In this paper, we study the problem of efficient dynamics modeling for multi-modal visual RL. We find that under the existence of modality heterogeneity, modality-correlated and distinct features are equally important but play different roles in reflecting the evolution of environmental dynamics. Motivated by this fact, we propose Dissected Dynamics Modeling (DDM), a novel multi-modal dynamics modeling method for visual RL. Unlike existing methods, DDM explicitly distinguishes consistent and inconsistent information across modalities and treats them separately with a divide-and-conquer strategy. This is done by dispatching the features carrying different information into distinct dynamics modeling pathways, which naturally form a series of implicit regularizations along the learning trajectories. In addition, a reward predictive function is further introduced to filter task-irrelevant information in both modality-consistent and inconsistent features, ensuring information integrity while avoiding potential distractions. Extensive experiments show that DDM consistently achieves competitive performance in challenging multi-modal visual environments.

Novel view synthesis under sparse views has been a long-term important challenge in 3D reconstruction. Existing works mainly rely on introducing external semantic or depth priors to supervise the optimization of 3D representations. However, the diffusion model, as an external prior that can directly provide visual supervision, has always underperformed in sparse-view 3D reconstruction using Score Distillation Sampling (SDS) due to the low information entropy of sparse views compared to text, leading to optimization challenges caused by mode deviation. To this end, we present a thorough analysis of SDS from the mode-seeking perspective and propose Inline Prior Guided Score Matching (IPSM), which leverages visual inline priors provided by pose relationships between viewpoints to rectify the rendered image distribution and decomposes the original optimization objective of SDS, thereby offering effective diffusion visual guidance without any fine-tuning or pre-training. Furthermore, we propose the IPSM-Gaussian pipeline, which adopts 3D Gaussian Splatting as the backbone and supplements depth and geometry consistency regularization based on IPSM to further improve inline priors and rectified distribution. Experimental results on different public datasets show that our method achieves state-of-the-art reconstruction quality. The code is released at https://github.com/iCVTEAM/IPSM.

Although substantial progress has been made in audiodriven talking video synthesis, there still remain two major difficulties: existing works 1) need a long sequence of training dataset (>1h) to synthesize co-speech gestures, which causes a significant limitation on their applicability; 2) usually fail to generate long sequences, or can only generate long sequences without enough diversity. To solve these challenges, we propose a Disentangled Recurrent Representation Learning framework to synthesize long diversified gesture sequences with a short training video of around 2 minutes. In our framework, we first make a disentangled latent space assumption to encourage unpaired audio and pose combinations, which results in diverse "one-to-many" mappings in pose generation. Next, we apply a recurrent inference module to feed back the last generation as initial guidance to the next phase, enhancing the long-term video generation of full continuity and diversity. Comprehensive experimental results verify that our model can generate realistic synchronized full-body talking videos with training data efficiency.

With the benefit of deep learning techniques, recent researches have made significant progress in image compression artifacts reduction. Despite their improved performances, prevailing methods only focus on learning a mapping from the compressed image to the original one but ignore the intrinsic attributes of the given compressed images, which greatly harms the performance of downstream parsing tasks. Different from these methods, we propose to decouple the intrinsic attributes into two complementary features for artifacts reduction, i.e., the compression-insensitive features to regularize the high-level semantic representations during training and the compression-sensitive features to be aware of the compression degree. To achieve this, we first employ adversarial training to regularize the compressed and original encoded features for retaining high-level semantics, and we then develop the compression quality-aware feature encoder for compression-sensitive features. Based on these dual complementary features, we propose a Dual Awareness Guidance Network (DAGN) to utilize these awareness features as transformation guidance during the decoding phase. In our proposed DAGN, we develop a cross-feature fusion module to maintain the consistency of compression-insensitive features by fusing compression-insensitive features into the artifacts reduction baseline. Our method achieves an average 2.06 dB PSNR gains on BSD500, outperforming state-of-the-art methods, and only requires 29.7 ms to process one image on BSD500. Besides, the experimental results on LIVE1 and LIU4K also demonstrate the efficiency, effectiveness, and superiority of the proposed method in terms of quantitative metrics, visual quality, and downstream machine vision tasks.

Few-shot class-incremental learning (FSCIL) aims at learning to classify new classes continually from limited samples without forgetting the old classes. The mainstream framework tackling FSCIL is first to adopt the cross-entropy (CE) loss for training at the base session, then freeze the feature extractor to adapt to new classes. However, in this work, we find that the CE loss is not ideal for the base session training as it suffers poor class separation in terms of representations, which further degrades generalization to novel classes. One tempting method to mitigate this problem is to apply an additional naïve supervised contrastive learning (SCL) in the base session. Unfortunately, we find that although SCL can create a slightly better representation separation among different base classes, it still struggles to separate base classes and new classes. Inspired by the observations made, we propose Semantic-Aware Virtual Contrastive model (SAVC), a novel method that facilitates separation between new classes and base classes by introducing virtual classes to SCL. These virtual classes, which are generated via pre-defined transformations, not only act as placeholders for unseen classes in the representation space, but also provide diverse semantic information. By learning to recognize and contrast in the fantasy space fostered by virtual classes, our SAVC significantly boosts base class separation and novel class generalization, achieving new state-of-the-art performance on the three widely-used FSCIL benchmark datasets. Code is available at: https://github.com/zysong0113/SAVC.

Efficient motion and appearance modeling are critical for vision-based Reinforcement Learning (RL). However, existing methods struggle to reconcile motion and appearance information within the state representations learned from a single observation encoder. To address the problem, we present Synergizing Interactive Motion-appearance Understanding (Simoun), a unified framework for vision-based RL Given consecutive observation frames, Simoun deliberately and interactively learns both motion and appearance features through a dual-path network architecture. The learning process collaborates with a structural interactive module, which explores the latent motion-appearance structures from the two network paths to leverage their complementarity. To promote sample efficiency, we further design a consistency-guided curiosity module to encourage the exploration of under-learned observations. During training, the curiosity module provides intrinsic rewards according to the consistency of environmental temporal dynamics, which are deduced from both motion and appearance network paths. Experiments conducted on Deep-Mind control suite and CARLA automatic driving benchmarks demonstrate the effectiveness of Simoun, where it performs favorably against state-of-the-art methods.

Vision-based reinforcement learning (RL) depends on discriminative representation encoders to abstract the observation states. Despite the great success of increasing CNN parameters for many supervised computer vision tasks, reinforcement learning with temporal-difference (TD) losses cannot benefit from it in most complex environments. In this paper, we analyze that the training instability arises from the oscillating self-overfitting of the heavy-optimizable encoder. We argue that serious oscillation will occur to the parameters when enforced to fit the sensitive TD targets, causing uncertain drifting of the latent state space and thus transmitting these perturbations to the policy learning. To alleviate this phenomenon, we propose a novel asymmetric interactive cooperation approach with the interaction between a heavy-optimizable encoder and a supportive light-optimizable encoder, in which both their advantages are integrated including the highly discriminative capability as well as the training stability. We also present a greedy bootstrapping optimization to isolate the visual perturbations from policy learning, where representation and policy are trained sufficiently by turns. Finally, we demonstrate the effectiveness of our method in utilizing larger visual models by first-person highway driving task CARLA and Vizdoom environments.

Integrating RGB frames with alternative modality inputs is gaining increasing traction in many vision-based reinforcement learning (RL) applications. Existing multi-modal vision-based RL methods usually follow a Global Value Estimation (GVE) pipeline, which uses a fused modality feature to obtain a unified global environmental description. However, such a feature-level fusion paradigm with a single critic may fall short in policy learning as it tends to overlook the distinct values of each modality. To remedy this, this paper proposes a Local modality-customized Value Estimation (LVE) paradigm, which dynamically estimates the contribution and adjusts the importance weight of each modality from a value-level perspective. Furthermore, a task-contextual re-fusion process is developed to achieve a task-level re-balance of estimations from both feature and value levels. To this end, a Hierarchical Adaptive Value Estimation (HAVE) framework is formed, which adaptively coordinates the contributions of individual modalities as well as their collective efficacy. Agents trained by HAVE are able to exploit the unique characteristics of various modalities while capturing their intricate interactions, achieving substantially improved performance. We specifically highlight the potency of our approach within the challenging landscape of autonomous driving, utilizing the CARLA benchmark with neuromorphic event and depth data to demonstrate HAVE's capability and the effectiveness of its distinct components.

Human part segmentation is a crucial but challenging task in computer vision. Recent works have achieved progress with the help of pixel-wise annotations. However, annotating pixel-wise masks especially at part-level is a tedious and labor-intensive procedure. To overcome this problem, we propose a part evolution framework to learn reliable predictions from weak pose annotations, which are much easier to collect. Our framework is composed of two essential modules: the first part adaptation module is designed to learn the deep prior knowledge from three related tasks, i.e., pose estimation, part-level and object-level segmentation; the second module is the part evolution module, which refines the part priors from deep predictions with the boundary-aware optimization algorithm. These two modules are conducted iteratively to evolve pose keypoint annotations into reliable part priors. Experimental evidence shows that our weakly-supervised approach generates comparable results with the state-of-the-art strongly-supervised methods on public benchmarks, and also validates the potential of notable improvements when combining weak labels with existing part segmentation masks.

Few-shot learning aims to recognize novel queries with limited support samples by learning from base knowledge. Recent progress in this setting assumes that the base knowledge and novel query samples are distributed in the same domains, which are usually infeasible for realistic applications. Toward this issue, we propose to address the cross-domain few-shot learning problem where only extremely few samples are available in target domains. Under this realistic setting, we focus on the fast adaptation capability of meta-learners by proposing an effective dual adaptive representation alignment approach. In our approach, a prototypical feature alignment is first proposed to recalibrate support instances as prototypes and reproject these prototypes with a differentiable closed-form solution. Therefore feature spaces of learned knowledge can be adaptively transformed to query spaces by the cross-instance and cross-prototype relations. Besides the feature alignment, we further present a normalized distribution alignment module, which exploits prior statistics of query samples for solving the covariant shifts among the support and query samples. With these two modules, a progressive meta-learning framework is constructed to perform the fast adaptation with extremely few-shot samples while maintaining its generalization capabilities. Experimental evidence demonstrates our approach achieves new state-of-the-art results on 4 CDFSL benchmarks and 4 fine-grained cross-domain benchmarks.

The sensitivity of deep neural networks to compressed images hinders their usage in many real applications, which means classification networks may fail just after taking a screenshot and saving it as a compressed file. In this paper, we argue that neglected disposable coding parameters stored in compressed files could be picked up to reduce the sensitivity of deep neural networks to compressed images. Specifically, we resort to using one of the representative parameters, quantization steps, to facilitate image classification. Firstly, based on quantization steps, we propose a novel quantization aware confidence (QAC), which is utilized as sample weights to reduce the influence of quantization on network training. Secondly, we utilize quantization steps to alleviate the variance of feature distributions, where a quantization aware batch normalization (QABN) is proposed to replace batch normalization of classification networks. Extensive experiments show that the proposed method significantly improves the performance of classification networks on CIFAR-10, CIFAR-100, and ImageNet.

Generalizable person re-identification aims to achieve a well generalization capability on target domains without accessing target data. Existing methods focus on suppressing domain-specific information or simulating unseen environments by meta-learning strategies, which could damage the capture ability on fine-grained visual patterns or lead to overfitting issues by the repetitive training of episodes. In this paper, we revisit the stochastic behaviors from two different perspectives: 1) Stochastic splitting-sliding sampler. It splits domain sources into approximately equal sample-size subsets and selects several subsets from various sources by a sliding window, forcing the model to step out of local minimums under stochastic sources. 2) Variance-varying gradient dropout. Gradients in parts of network are also selected by a sliding window and multiplied by binary masks generated from Bernoulli distribution, making gradients in varying variance and preventing the model from local minimums. By applying these two proposed stochastic behaviors, the model achieves a better generalization performance on unseen target domains without any additional computation costs or auxiliary modules. Extensive experiments demonstrate that our proposed model is effective and outperforms state-of-the-art methods on public domain generalizable person Re-ID benchmarks.

Generalization in image-based reinforcement learning (RL) aims to learn a robust policy that could be applied directly on unseen visual environments, which is a challenging task since agents usually tend to overfit to their training environment. To handle this problem, a natural approach is to increase the data diversity by image based augmentations. However, different with most vision tasks such as classification and detection, RL tasks are not always invariant to spatial based augmentations due to the entanglement of environment dynamics and visual appearance. In this paper, we argue with two principles for augmentations in RL: First, the augmented observations should facilitate learning a universal policy, which is robust to various distribution shifts. Second, the augmented data should be invariant to the learning signals such as action and reward. Following these rules, we revisit image-based RL tasks from the view of frequency domain and propose a novel augmentation method, namely Spectrum Random Masking (SRM),which is able to help agents to learn the whole frequency spectrum of observation for coping with various distributions and compatible with the pre-collected action and reward corresponding to original observation. Extensive experiments conducted on DMControl Generalization Benchmark demonstrate the proposed SRM achieves the state-of-the-art performance with strong generalization potentials.

Depth information can benefit various computer vision tasks on both images and videos. However, depth maps may suffer from invalid values in many pixels, and also large holes. To improve such data, we propose a joint self-supervised and reference-guided learning approach for depth inpainting. For the self-supervised learning strategy, we introduce an improved spatial convolutional sparse coding module in which total variation regularization is employed to enhance the structural information while preserving edge information. This module alternately learns a convolutional dictionary and sparse coding from a corrupted depth map. Then, both the learned convolutional dictionary and sparse coding are convolved to yield an initial depth map, which is effectively smoothed using local contextual information. The reference-guided learning part is inspired by the fact that adjacent pixels with close colors in the RGB image tend to have similar depth values. We thus construct a hierarchical joint bilateral filter module using the corresponding color image to fill in large holes. In summary, our approach integrates a convolutional sparse coding module to preserve local contextual information and a hierarchical joint bilateral filter module for filling using specific adjacent information. Experimental results show that the proposed approach works well for both invalid value restoration and large hole inpainting.

Fine-grained object recognition aims to learn effective features that can identify the subtle differences between visually similar objects. Most of the existing works tend to amplify discriminative part regions with attention mechanisms. Besides its unstable performance under complex backgrounds, the intrinsic interrelationship between different semantic features is less explored. Toward this end, we propose an effective graph-based relation discovery approach to build a contextual understanding of high-order relationships. In our approach, a high-dimensional feature bank is first formed and jointly regularized with semantic- and positional-aware high-order constraints, endowing rich attributes to feature representations. Second, to overcome the high-dimension curse, we propose a graph-based semantic grouping strategy to embed this high-order tensor bank into a low-dimensional space. Meanwhile, a group-wise learning strategy is proposed to regularize the features focusing on the cluster embedding center. With the collaborative learning of three modules, our module is able to grasp the stronger contextual details of fine-grained objects. Experimental evidence demonstrates our approach achieves new state-of-the-art on 4 widely-used fine-grained object recognition benchmarks.

In this paper, we propose a talking face generation method that takes an audio signal as input and a short target video clip as reference, and synthesizes a photo-realistic video of the target face with natural lip motions, head poses, and eye blinks that are in-sync with the input audio signal. We note that the synthetic face attributes include not only explicit ones such as lip motions that have high correlations with speech, but also implicit ones such as head poses and eye blinks that have only weak correlation with the input audio. To model such complicated relationships among different face attributes with input audio, we propose a FACe Implicit Attribute Learning Generative Adversarial Network (FACIAL-GAN), which integrates the phonetics-aware, context-aware, and identity-aware information to synthesize the 3D face animation with realistic motions of lips, head poses, and eye blinks. Then, our Rendering-to-Video network takes the rendered face images and the attention map of eye blinks as input to generate the photorealistic output video frames. Experimental results and user studies show our method can generate realistic talking face videos with not only synchronized lip motions, but also natural head movements and eye blinks, with better qualities than the results of state-of-the-art methods.

The simultaneous recognition of multiple objects in one image remains a challenging task, spanning multiple events in the recognition field such as various object scales, inconsistent appearances, and confused inter-class relationships. Recent research efforts mainly resort to the statistic label co-occurrences and linguistic word embedding to enhance the unclear semantics. Different from these researches, in this paper, we propose a novel Transformer-based Dual Relation learning framework, constructing complementary relationships by exploring two aspects of correlation, i.e., structural relation graph and semantic relation graph. The structural relation graph aims to capture long-range correlations from object context, by developing a cross-scale transformer-based architecture. The semantic graph dynamically models the semantic meanings of image objects with explicit semantic-aware constraints. In addition, we also incorporate the learnt structural relationship into the semantic graph, constructing a joint relation graph for robust representations. With the collaborative learning of these two effective relation graphs, our approach achieves new state-of-the-art on two popular multi-label recognition benchmarks, i.e. MS-COCO and VOC 2007 dataset.

The crucial problem in vehicle re-identification is to find the same vehicle identity when reviewing this object from cross-view cameras, which sets a higher demand for learning viewpoint-invariant representations. In this paper, we propose to solve this problem from two aspects: constructing robust feature representations and proposing camera-sensitive evaluations. We first propose a novel Heterogeneous Relational Complement Network (HRCN) by incorporating region-specific features and cross-level features as complements for the original high-level output. Considering the distributional differences and semantic misalignment, we propose graph-based relation modules to embed these heterogeneous features into one unified high-dimensional space. On the other hand, considering the deficiencies of cross-camera evaluations in existing measures (i.e., CMC and AP), we then propose a Cross-camera Generalization Measure (CGM) to improve the evaluations by introducing position-sensitivity and cross-camera generalization penalties. We further construct a new benchmark of existing models with our proposed CGM and experimental results reveal that our proposed HRCN model achieves new state-of-the-art in VeRi-776, VehicleID, and VERI-Wild.

Few-shot learning aims to recognize the novel categories from a few examples. However, most of the existing approaches usually focus on general image classification and fail to handle subtle differences between images. To alleviate this issue, we propose a trilinear spatial-awareness network for few-shot-grained visual recognition, called S3Net, which is composed of a spatial selection module, structural pyramid descriptor, and subtle difference mining module. Specifically, we first build the global relation to strengthen the features by spatial selection module. The structural pyramid descriptor then constructs a multi-scale representation for enhancing the rich contextual information by exploiting different receptive fields in the same feature layer. Furthermore, a similarity loss based on local descriptors and a global classification loss is design to help the network learn discrimination capability by handling subtle differences in confused or near-duplicated samples. Extensive experiments on 4 few-shot fine-grained benchmarks demonstrate that our proposed approach is effective and outperforms state-of-the-art models by large margins.

Person Re-Identification (Re-Id) in occlusion scenarios is a challenging problem because a pedestrian can be partially occluded. The use of local information for feature extraction and matching is still necessary. Therefore, we propose a Pose-guided inter- and intra-part relational transformer (Pirt) for occluded person Re-Id, which builds part-aware long-term correlations by introducing transformer. In our framework, we firstly develop a pose-guided feature extraction module with regional grouping and mask construction for robust feature representations. The positions of a pedestrian in the image under surveillance scenarios are relatively fixed, hence we propose intra-part and inter-part relational transformer. The intra-part module creates local relations with mask-guided features, while the inter-part relationship builds correlations with transformers, to develop cross relationships between part nodes. With the collaborative learning inter- and intra-part relationships, experiments reveal that our proposed Pirt model achieves a new state of the art on the public occluded dataset, and further extensions on standard non-occluded person Re-Id datasets also reveal our comparable performances.

Multi-label image recognition aims to recognize multiple objects simultaneously in one image. Recent ideas to solve this problem have focused on learning dependencies of label co-occurrences to enhance the high-level semantic representations. However, these methods usually neglect the important relations of intrinsic visual structures and face difficulties in understanding contextual relationships. To build the global scope of visual context as well as interactions between visual modality and linguistic modality, we propose the Multi-Modal Multi-label recognition TRansformers (M3TR) with the ternary relationship learning for inter-and intra-modalities. For the intra-modal relationship, we make insightful conjunction of CNNs and Transformers, which embeds visual structures into the high-level features by learning the semantic cross-attention. For constructing the interactions between the visual and linguistic modalities, we propose a linguistic cross-attention to embed the class-wise linguistic information into the visual structure learning, and finally present a linguistic guided enhancement module to enhance the representation of high-level semantics. Experimental evidence reveals that with the collaborative learning of ternary relationship, our proposed M3TR achieves new state-of-the-art on two public multi-label recognition benchmarks.

Semantic object part segmentation is a fundamental task in object understanding and geometric analysis. The clear understanding of part relationships can be of great use to the segmentation process. In this work, we propose a novel Ordinal Multi-task Part Segmentation (OMPS) approach which explicitly models the part ordinal relationship to guide the segmentation process in a recurrent manner. Quantitative and qualitative experiments are conducted first to explore the mutual impacts among object parts and then an ordinal part inference algorithm is formulated via experimental observations. Specifically, our framework is mainly composed of two modules, the forward module to segment multiple parts as individual subtasks with prior knowledge, and the recurrent module to generate appropriate part priors with the ordinal inference algorithm. These two modules work iteratively to optimize the segmentation performance and the network parameters. Experimental results show that our approach outperforms the state-of-the-art models on human and vehicle part parsing benchmarks. Comprehensive evaluations are conducted to demonstrate the effectiveness of our approach in object part segmentation.

Close your eyes and listen to music, one can easily imagine an actor dancing rhythmically along with the music. These dance movements are usually made up of dance movements you have seen before. In this paper, we propose to reproduce such an inherent capability of the human-being within a computer vision system. The proposed system consists of three modules. To explore the relationship between music and dance movements, we propose a cross-modal alignment module that focuses on dancing video clips, accompanied by pre-designed music, to learn a system that can judge the consistency between the visual features of pose sequences and the acoustic features of music. The learned model is then used in the imagination module to select a pose sequence for the given music. Such pose sequence selected from the music, however, is usually discontinuous. To solve this problem, in the spatial-temporal alignment module we develop a spatial alignment algorithm based on the tendency and periodicity of dance movements to predict dance movements between discontinuous fragments. In addition, the selected pose sequence is often misaligned with the music beat. To solve this problem, we further develop a temporal alignment algorithm to align the rhythm of music and dance. Finally, the processed pose sequence is used to synthesize realistic dancing videos in the imagination module. The generated dancing videos match the content and rhythm of the music. Experimental results and subjective evaluations show that the proposed approach can perform the function of generating promising dancing videos by inputting music.

Fine-grained visual recognition is to classify objects with visually similar appearances into subcategories, which has made great progress with the development of deep CNNs. However, handling subtle differences between different subcategories still remains a challenge. In this paper, we propose to solve this issue in one unified framework from two aspects, i.e., constructing feature-level interrelationships, and capturing part-level discriminative features. This framework, namely PArt-guided Relational Transformers (PART), is proposed to learn the discriminative part features with an automatic part discovery module, and to explore the intrinsic correlations with a feature transformation module by adapting the Transformer models from the field of natural language processing. The part discovery module efficiently discovers the discriminative regions which are highly-corresponded to the gradient descent procedure. Then the second feature transformation module builds correlations within the global embedding and multiple part embedding, enhancing spatial interactions among semantic pixels. Moreover, our proposed approach does not rely on additional part branches in the inference time and reaches state-of-the-art performance on 3 widely-used fine-grained object recognition benchmarks. Experimental results and explainable visualizations demonstrate the effectiveness of our proposed approach.

Conventional RGB-D salient object detection methods aim to leverage depth as complementary information to find the salient regions in both modalities. However, the salient object detection results heavily rely on the quality of captured depth data which sometimes are unavailable. In this work, we make the first attempt to solve the RGB-D salient object detection problem with a novel depth-awareness framework. This framework only relies on RGB data in the testing phase, utilizing captured depth data as supervision for representation learning. To construct our framework as well as achieving accurate salient detection results, we propose a Ubiquitous Target Awareness (UTA) network to solve three important challenges in RGB-D SOD task: 1) a depth awareness module to excavate depth information and to mine ambiguous regions via adaptive depth-error weights, 2) a spatial-aware cross-modal interaction and a channel-aware cross-level interaction, exploiting the low-level boundary cues and amplifying high-level salient channels, and 3) a gated multi-scale predictor module to perceive the object saliency in different contextual scales. Besides its high performance, our proposed UTA network is depth-free for inference and runs in real-time with 43 FPS. Experimental evidence demonstrates that our proposed network not only surpasses the state-of-the-art methods on five public RGB-D SOD benchmarks by a large margin, but also verifies its extensibility on five public RGB SOD benchmarks.

Understanding an image with 3D representations has been an increasingly attractive topic in computer vision. The state-of the-art 3D reconstruction methods usually focus on the reconstruction of the holistic object, while missing important part information, which is crucial in robotic interaction and virtual reality applications. To solve this issue, we make the first attempt to reconstruct the 3D models with part-level representations in a unified framework. With the input of the singleview images, we first develop a feature enhancement encoder to incorporate discriminative local features into the feature representation. The local features are selected adaptively by a learnable local awareness module. Then the enhanced local features are fused with the global branch to form the 3D representations. We then develop a 3D part generator to decode the image priors to 3D parts with a 3D focal loss, which enables the representations of small parts. Experimental results indicate that our model generates reliable part-level structures while achieving state-of-the-art performance in object-level recovering.

Given base classes with sufficient labeled samples, the target of few-shot classification is to recognize unlabeled samples of novel classes with only a few labeled samples. Most existing methods only pay attention to the relationship between labeled and unlabeled samples of novel classes, which do not make full use of information within base classes. In this paper, we make two contributions to investigate the few-shot classification problem. First, we report a simple and effective baseline trained on base classes in the way of traditional supervised learning, which can achieve comparable results to the state of the art. Second, based on the baseline, we propose a cooperative bi-path metric for classification, which leverages the correlations between base classes and novel classes to further improve the accuracy. Experiments on two widely used benchmarks show that our method is a simple and effective framework, and a new state of the art is established in the few-shot classification field.

Salient object detection (SOD) is a crucial and preliminary task for many computer vision applications, which have made progress with deep CNNs. Most of the existing methods mainly rely on the RGB information to distinguish the salient objects, which faces difficulties in some complex scenarios. To solve this, many recent RGBD-based networks are proposed by adopting the depth map as an independent input and fuse the features with RGB information. Taking the advantages of RGB and RGBD methods, we propose a novel depth-aware salient object detection framework, which has following superior designs: 1) It does not rely on depth data in the testing phase. 2) It comprehensively optimizes SOD features with multi-level depth-aware regularizations. 3) The depth information also serves as error-weighted map to correct the segmentation process. With these insightful designs combined, we make the first attempt in realizing an unified depth-aware framework with only RGB information as input for inference, which not only surpasses the state-of-the-art performance on five public RGB SOD benchmarks, but also surpasses the RGBD-based methods on five benchmarks by a large margin, while adopting less information and implementation light-weighted.

Recent years have witnessed increasing attention in cartoon media, powered by the strong demands of industrial applications. As the first step to understand this media, cartoon face recognition is a crucial but less-explored task with few datasets proposed. In this work, we first present a new challenging benchmark dataset, consisting of 389,678 images of 5,013 cartoon characters annotated with identity, bounding box, pose, and other auxiliary attributes. The dataset, named iCartoonFace, is currently the largest-scale, high-quality, rich-annotated, and spanning multiple occurrences in the field of image recognition, including near-duplications, occlusions, and appearance changes. In addition, we provide two types of annotations for cartoon media, i.e., face recognition, and face detection, with the help of a semi-automatic labeling algorithm. To further investigate this challenging dataset, we propose a multi-task domain adaptation approach that jointly utilizes the human and cartoon domain knowledge with three discriminative regularizations. We hence perform a benchmark analysis of the proposed dataset and verify the superiority of the proposed approach in the cartoon face recognition task. The dataset is available at https://iqiyi.cn/icartoonface.

Vehicle re-identification (Re-ID) has been attracting more interests in computer vision owing to its great contributions in urban surveillance and intelligent transportation. With the development of deep learning approaches, vehicle Re-ID still faces a near-duplicate challenge, which is to distinguish different instances with nearly identical appearances. Previous methods simply rely on the global visual features to handle this problem. In this paper, we proposed a simple but efficient part-regularized discriminative feature preserving method which enhances the perceptive ability of subtle discrepancies. We further develop a novel framework to integrate part constrains with the global Re-ID modules by introducing an detection branch. Our framework is trained end-to-end with combined local and global constrains. Specially, without the part-regularized local constrains in inference step, our Re-ID network outperforms the state-of-the-art method by a large margin on large benchmark datasets VehicleID and VeRi-776.

Object part parsing in the wild, which requires to simultaneously detect multiple object classes in the scene and accurately segments semantic parts within each class, is challenging for the joint presence of class-level and part-level ambiguities. Despite its importance, however, this problem is not sufficiently explored in existing works. In this paper, we propose a joint parsing framework with boundary and semantic awareness to address this challenging problem. To handle part-level ambiguity, a boundary awareness module is proposed to make mid-level features at multiple scales attend to part boundaries for accurate part localization, which are then fused with high-level features for effective part recognition. For class-level ambiguity, we further present a semantic awareness module that selects discriminative part features relevant to a category to prevent irrelevant features being merged together. The proposed modules are lightweight and implementation friendly, improving the performance substantially when plugged into various baseline architectures. Without bells and whistles, the full model sets new state-of-the-art results on the Pascal-Part dataset, in both multi-class and the conventional single-class setting, while running substantially faster than recent high-performance approaches.

Along with the development of virtual reality (VR), omnidirectional images play an important role in producing multimedia content with an immersive experience. However, despite various existing approaches for omnidirectional image stitching, how to quantitatively assess the quality of stitched images is still insufficiently explored. To address this problem, we first establish a novel omnidirectional image dataset containing stitched images as well as dual-fisheye images captured from standard quarters of 0$^\circ$, 90$^\circ$, 180$^\circ$, and 270$^\circ$. In this manner, when evaluating the quality of an image stitched from a pair of fisheye images (\eg, 0$^\circ$ and 180$^\circ$), the other pair of fisheye images (\eg, 90$^\circ$ and 270$^\circ$) can be used as the cross-reference to provide ground-truth observations of the stitching regions. Based on this dataset, we propose a set of Omnidirectional Stitching Image Quality Assessment (OS-IQA) metrics. In these metrics, the stitching regions are assessed by exploring the local relationships between the stitched image and its cross-reference with histogram statistics, perceptual hash and sparse reconstruction, while the whole stitched images are assessed by the global indicators of color difference and fitness of blind zones.Qualitative and quantitative experiments show our method outperforms the classic IQA metrics and is highly consistent with human subjective evaluations. To the best of our knowledge, it is the first attempt that assesses the stitching quality of omnidirectional images by using cross-references.