Reliable zero-shot detection of out-of-distribution (OOD) inputs is critical for deploying vision-language models in open-world settings. However, the lack of labeled negatives in zero-shot OOD detection necessitates proxy signals that remain effective under distribution shift. Existing negative-label methods rely on a fixed set of textual proxies, which (i) sparsely sample the semantic space beyond in-distribution (ID) classes and (ii) remain static while only visual features drift, leading to cross-modal misalignment and unstable predictions. In this paper, we propose CoEvo, a training- and annotation-free test-time framework that performs bidirectional, sample-conditioned adaptation of both textual and visual proxies. Specifically, CoEvo introduces a proxy-aligned co-evolution mechanism to maintain two evolving proxy caches, which dynamically mines contextual textual negatives guided by test images and iteratively refines visual proxies, progressively realigning cross-modal similarities and enlarging local OOD margins. Finally, we dynamically re-weight the contributions of dual-modal proxies to obtain a calibrated OOD score that is robust to distribution shift. Extensive experiments on standard benchmarks demonstrate that CoEvo achieves state-of-the-art performance, improving AUROC by 1.33% and reducing FPR95 by 45.98% on ImageNet-1K compared to strong negative-label baselines.

We propose InterFold, a framework for learning and applying interpretable semantic manifolds in latent diffusion models, without requiring binary or paired supervision. Existing methods for semantic editing either rely on limited paired data or uncover only coarse, unsupervised directions that fail to capture user-specific, fine-grained attributes. InterFold addresses these limitations by learning a target attribute manifold in the H-space of diffusion models using only a set of positive, unlabeled examples. To edit a new image, InterFold projects its H-space representation toward this learned manifold through test-time optimization, enabling precise, identity-preserving modifications of complex, non-binary concepts. To make these edits effective in modern latent diffusion models, we introduce the Manifold Adapter, a lightweight cross-attention module that transfers semantic intent from edited H-space codes into the generative latent space, without altering the pretrained model. Extensive experiments demonstrate that InterFold achieves superior edit accuracy and identity consistency compared to existing methods, offering a flexible and interpretable solution for high-fidelity semantic image editing.

Generating high-quality sketches from video requires a nuanced understanding of semantic content and visual structure, particularly for complex scenes across diverse sketch styles. Efficient and flexible video-to-sketch style transformation remains a significant challenge. We introduce VideoSketcher, a training-free framework for style-controllable sketch video generation that preserves frame structure while applying specified sketch aesthetics. Leveraging text-to-image diffusion models, VideoSketcher utilizes strong semantic priors without the need for extensive training. Our approach enforces temporal consistency by retaining latent information across frames and employs a Time-Linked Attention mechanism to capture structural elements from the source video and inject stylistic information from the reference image. To bridge the semantic gap between sketches and original video content, we introduce Sketch Directive Amplification for selective transfer of stylistic features. Additionally, a Stroke Graph Regularization strategy, comprising line and point loss, refines line consistency in the latent space. Extensive experiments validate VideoSketcher’s superior temporal stability and fidelity across diverse sketch styles and content. Video demos can be found in the supplementary materials.

Translating real portrait video into anime is an application of interest to both consumers and researchers. However, anime differs considerably from portraits, making portrait-to-anime translation challenging. Existing StyleGAN-based portrait stylization works assume that the portrait and stylized generators share the same latent space, but this assumption fails in the style of anime due to the large domain gap. Moreover, directly applying them to each video frame often leads to undesirable temporal inconsistencies. In this paper, we argue that two latent spaces with a large domain gap cannot be shared but can be related by a transformation, and develop a cyclic transformation network to connect the two spaces with two cycle constraints. This provides high-quality translation for each frame. We extend our framework to video transformation by proposing a novel frame interpolation constraint which ensures that in-between frames can be interpolated from their neighboring frames, guaranteeing temporal coherence across translated frames. Together with latent code smoothing regularization, this provides temporally coherent video-to-anime translation. Extensive experiments demonstrate that our framework outperforms state-of-the-art methods both qualitatively and quantitatively.

Long-term motion generation is a challenging task that requires producing coherent and realistic sequences over extended durations. Current methods primarily rely on framewise motion representations, which capture only static spatial details and overlook temporal dynamics. This approach leads to significant redundancy across the temporal dimension, complicating the generation of effective long-term motion. To overcome these limitations, we introduce the novel concept of Lagrangian Motion Fields, specifically designed for long-term motion generation. By treating each joint as a Lagrangian particle with uniform velocity over short intervals, our approach condenses motion representations into a series of "supermotions" (analogous to superpixels). This method seamlessly integrates static spatial information with interpretable temporal dynamics, transcending the limitations of existing network architectures and motion sequence content types. Our solution is versatile and lightweight, eliminating the need for neural network preprocessing. Our approach excels in tasks such as long-term music-to-dance generation and text-to-motion generation, offering enhanced efficiency, superior generation quality, and greater diversity compared to existing methods. Additionally, the adaptability of Lagrangian Motion Fields extends to applications like infinite motion looping and fine-grained controlled motion generation, highlighting its broad utility.

Effective segmentation of unseen categories in zero-shot semantic segmentation is hindered by models’ limited ability to interpret edges in unfamiliar contexts. In this paper, we propose EdgeCLIP, which addresses this by integrating CLIP with explicit edge-awareness. Based on the premise that edge variation patterns are similar across both seen and unseen class objects, EdgeCLIP introduces the Contextual Edge Sensing module. This module accurately discerns and utilizes edge information, which is crucial in complex border areas where conventional models struggle. Further, our Text-Guided Dense Feature Matching strategy precisely aligns text encodings with corresponding visual edge features, effectively distinguishing them from background edges. This strategy not only optimizes the training of CLIP’s image and text encoders but also leverages the intrinsic completeness of objects, enhancing the model’s ability to generalize and accurately segment objects in unseen classes. EdgeCLIP significantly outperforms the current state-of-the-art method, achieving a deep impressive margin of 17.5% on COCO-20i datasets. Our code is available at github.com/aqingaqinghh/EdgeCLIP.

Monocular 3D human pose estimation remains a challenging task due to inherent depth ambiguities and occlusions. Compared to traditional methods based on Transformers or Convolutional Neural Networks (CNNs), recent diffusion-based approaches have shown superior performance, leveraging their probabilistic nature and high-fidelity generation capabilities. However, these methods often fail to account for the spatial and temporal correlations across predicted frames, resulting in limited temporal consistency and inferior accuracy in predicted 3D pose sequences. To address these shortcomings, this paper proposesStarPose, an autoregressive diffusion framework that effectively incorporates historical 3D pose predictions and spatial-temporal physical guidance to significantly enhance both the accuracy and temporal coherence of pose predictions. Unlike existing approaches,StarPosemodels the 2D-to-3D pose mapping as an autoregressive diffusion process. By synergically integrating previously predicted 3D poses with 2D pose inputs via a Historical Pose Integration Module (HPIM), the framework generates rich and informative historical pose embeddings that guide subsequent denoising steps, ensuring temporally consistent predictions. In addition, a fully plug-and-play Spatial-Temporal Physical Guidance (STPG) mechanism is tailored to refine the denoising process in an iterative manner, which further enforces spatial anatomical plausibility and temporal motion dynamics, rendering robust and realistic pose estimates. Extensive experiments on benchmark datasets demonstrate thatStarPoseoutperforms state-of-the-art methods, achieving superior accuracy and temporal consistency in 3D human pose estimation. Code is available at https://github.com/wileychan/StarPose.

The goal of point cloud salient object detection is to identify and segment the most prominent areas or objects within a 3D point cloud. Research on point cloud SOD is still in its early stages, and many existing methods fail to fully utilize the rich geometric information inherent in point clouds. To address this limitation, we introduce GeoEdgeFormer, an effective Edge-Enhanced Transformer Network, designed specif ically for 3D salient object detection. GeoEdgeFormer employs an encoder-decoder architecture featuring two novel components: the Residual Edge Convolution (REC) and the Global Contextual Transformer (GCT). In the encoder, we propose the REC, which is designed to maintain permutation invariance while capturing local geometric information. This component not only improves the model's ability to process complex point cloud data but also enhances its efficiency, making it suitable for dynamic. In the decoder, we introduce the GCT to learn scene-level contextual representations. The GCT integrates global semantics and multi-level features from the encoder into a cohesive global scene context. By effectively combining features from local and global levels, the model achieves a more comprehensive understanding of the scene's semantics, thereby enhancing its generalization ability. Extensive experiments on the PCSOD saliency dataset demonstrate that our proposed GeoEdgeFormer achieves state of-the-art performance.

Weakly supervised text-based person re-identification (Text-ReID) confronts the challenge of matching target person images with textual descriptions, hindered by the absence of identity annotations during training. Traditional approaches, which rely solely on global features, overlook the rich, fine-grained information within both text and image modalities. Besides, merely aligning features at the semantic level is insufficient due to the significant differences in feature representation spaces between the two modalities. Existing methods also neglect the information inequality caused by person-irrelevant factors in images. In this paper, we introduce a novel framework called Attribute-Centric Cross-modal Alignment (ACCA), specifically designed to overcome these issues. Our approach concentrates on two main aspects: visual-text attribute alignment and prediction distribution alignment. To effectively capture fine-grained information without identity labels, we implement a visual-text attribute alignment method based on momentum contrastive learning to synchronize visual and textual attribute features within a unified embedding space. We also propose a unique strategy for negative sample filtering and enrichment, creating robust and comprehensive negative attribute sample spaces to support the attribute alignment. Additionally, we establish two methods of label-free prediction distribution alignment to encourage the learning of invariant feature representations across modalities. The first method, bias-reduction distribution alignment, aligns features and predictions within each text-image pair by utilizing semantic information from the text and reduces the impact of person-irrelevant factors in images. The second method, global-attribute distribution alignment, enhances the interaction between global and local prediction distributions across visual and textual modalities. Extensive experiments on CUHK-PEDES, ICFG-PEDES, and RSTPReid datasets validate our superior performances across all standard benchmarks.

Sketches, as a new solution in multimedia systems that can replace natural language, are characterized by sparse visual cues such as simple strokes that differ significantly from natural images containing complex elements such as background, foreground, and texture. This misalignment poses substantial challenges for zero-shot sketch-based image retrieval (ZS-SBIR). Prior approaches match sketches to full images and tend to overlook redundant elements in natural images, leading to model distraction and semantic ambiguity. To address this issue, we introduce a distraction-agnostic framework, purified cross-domain matching (PuXIM), which operates on a straightforward principle: masking and matching. We devise a visual-cross-linguistic (VxL) sampler that generates linguistic masks based on semantic labels to obscure semantically irrelevant image features. Our novel contribution is the concept of purified masked matching (PMM), which comprises two processes: (1)reconstruction, which compels the image encoder to reconstruct the masked image feature, and (2)interaction, which involves a transformer decoder that processes both sketch and masked image features to investigate cross-domain relationships for effective matching. Evaluated on the TU-Berlin, Sketchy, and QuickDraw datasets, PuXIM sets new benchmarks in terms of performance. Importantly, the distraction-agnostic nature of the matching process renders PuXIM more conducive to training, enabling efficient adaptation to zero-shot scenarios with reduced data requirements and low data quality.

Predicting 3D Semantic Scene Graphs (3DSSG) is vital for understanding complex scenes by constructing structured representations. Current methods struggle with significant granularity discrepancies among instances, often relying on features at a single scale, which hampers their ability to perceive and interact with differently sized instances. To tackle this challenge, we introduce GranSSG, a novel approach that integrates volumetric granular awareness into 3DSSG prediction. Central to GranSSG is the Volumetric Pooling block, which aggregates features from multiple instance volumes, enhancing the representation of instance patterns across different granularities. Complementing this, the Granularity Transformer block dynamically directs attention to instance features across various network layers, ensuring precise perception of instances regardless of their granularity. Furthermore, the Cross-Granularity Correlation Transformer block mitigates performance degradation in instance pair relationship prediction by adaptively fusing hybrid features from different granularities, providing a comprehensive representation of instance pairs. Extensive evaluations on the challenging 3DSSG benchmark demonstrate that GranSSG significantly enhances prediction performance, setting a new state-of-the-art in 3DSSG prediction.

Singing-driven 3D head animation is a compelling yet underexplored task with broad applications in virtual avatars, entertainment, and education. Existing speech-driven approaches, which typically map audio directly to motion through implicit phoneme-to-viseme correspondences, often yield over-smoothed, emotionally flat, and semantically inconsistent results. These limitations render them inadequate for the unique demands of singing-driven animation. To address this challenge, we propose Think2Sing, a unified diffusion-based framework that integrates pretrained large language models to generate semantically consistent and temporally coherent 3D head animations conditioned on both lyrics and acoustics. Central to our framework is the introduction of motion subtitles, a structured, time-aligned representation generated via a Singing Chain-of-Thought process with acoustic-guided retrieval. These subtitles provide region-specific expressive cues that serve as interpretable priors for animation synthesis. We further formulate head animation as motion intensity prediction over key facial regions, enabling fine-grained control and more faithful expressive modeling. To support this paradigm, we construct the first multimodal singing dataset with synchronized 3D motion, acoustic descriptors, and aligned motion subtitles, enabling semantically grounded and expressive motion learning. Extensive experiments demonstrate that Think2Sing significantly outperforms state-of-the-art methods in realism, expressiveness, and emotional fidelity. Furthermore, our framework supports flexible subtitle-conditioned editing, enabling precise and user-controllable animation synthesis.

We propose AvatarVTON, the first 4D virtual try-on framework that generates realistic try-on results from a single in-shop garment image, enabling free pose control, novel-view rendering, and diverse garment choices. Unlike existing methods, AvatarVTON supports dynamic garment interactions under single-view supervision, without relying on multi-view garment captures or physics priors. The framework consists of two key modules: (1) a Reciprocal Flow Rectifier, an optical-flow-based correction strategy without external priors that stabilizes avatar fitting and ensures temporal coherence; and (2) a Non-Linear Deformer, which decomposes Gaussian maps into view-pose-invariant and view-pose-specific components, enabling adaptive, non-linear garment deformations. To establish a benchmark for 4D virtual try-on, we extend existing baselines with unified modules for fair qualitative and quantitative comparisons. Extensive experiments show that AvatarVTON achieves high fidelity, diversity, and dynamic garment realism, making it well-suited for AR/VR, gaming, and digital-human applications.

With the revolution of generative AI, video-related tasks have been widely studied. However, current state-of-the-art video models still lag behind image models in visual quality and user control over generated content. In this paper, we introduce TokenWarping, a novel framework for temporally coherent video translation. Existing diffusion-based video editing approaches rely solely on key and value patches in self-attention to ensure temporal consistency, often sacrificing the preservation of local and structural regions. Critically, these methods overlook the significance of the query patches in achieving accurate feature aggregation and temporal coherence. In contrast, TokenWarping leverages complementary token priors by constructing temporal correlations across different frames. Our method begins by extracting optical flows from source videos. During the denoising process of the diffusion model, these optical flows are used to warp the previous frame's query, key, and value patches, aligning them with the current frame's patches. By directly warping the query patches, we enhance feature aggregation in self-attention, while warping the key and value patches ensures temporal consistency across frames. This token warping imposes explicit constraints on the self-attention layer outputs, effectively ensuring temporally coherent translation. Our framework does not require any additional training or fine-tuning and can be seamlessly integrated with existing text-to-image editing methods. We conduct extensive experiments on various video translation tasks, demonstrating that TokenWarping surpasses state-of-the-art methods both qualitatively and quantitatively. Video demonstrations are available in supplementary materials.

Tongue diagnosis is a vital tool in both Western and Traditional Chinese Medicine, providing key insights into a patient's health by analyzing tongue attributes. The COVID-19 pandemic has heightened the need for accurate remote medical assessments, emphasizing the importance of precise tongue attribute recognition via telehealth. To address this, we propose a Sign-Oriented multi-label Attributes Detection Framework. Our approach begins with an adaptive tongue feature extraction module that standardizes tongue images and mitigates environmental factors. This is followed by a Sign-oriented Network (SignNet) that identifies specific tongue attributes, emulating the diagnostic process of experienced practitioners and enabling comprehensive health evaluations. To validate our methodology, we developed an extensive tongue image dataset specifically designed for telemedicine. Unlike existing datasets, ours is tailored for remote diagnosis, with a comprehensive set of attribute labels. This dataset will be openly available, providing a valuable resource for research. Initial tests have shown improved accuracy in detecting various tongue attributes, highlighting our framework's potential as an essential tool for remote medical assessments.

Talking head video generation involves animating a still face image using facial motion cues derived from a driving video to replicate target poses and expressions. Traditional methods often rely on the assumption that the relative positions of facial keypoints remain unchanged. However, this assumption fails when keypoints are occluded or when the head is in a profile pose, leading to inconsistencies in identity and blurring in certain facial regions. In this paper, we introduce Occlusion-Insensitive Talking Head Video Generation, a novel approach that eliminates the reliance on spatial correlation of keypoints and instead leverages semantic correlation. Our method transforms facial features into a facelet semantic bank, where each facelet token represents a specific facial semantic. This bank is devoid of spatial information, allowing it to compensate for any invisible or occluded face regions during motion warping. The facelet compensation module then populates the facelet tokens within the initially warped features by learning a correlation matrix between facial semantics and the facelet bank. This approach enables precise compensation for occlusions and pose changes, enhancing the fidelity of the generated videos. Extensive experiments demonstrate that our method achieves state-of-the-art results, preserving source identity, maintaining fine-grained facial details, and capturing nuanced facial expressions with remarkable accuracy.

Personalized image generation has made significant strides in adapting content to novel concepts. However, a persistent challenge remains: balancing the accurate reconstruction of unseen concepts with the need for editability according to the prompt, especially when dealing with the complex nuances of facial features. In this study, we delve into the temporal dynamics of the text-to-image conditioning process, emphasizing the crucial role of stage partitioning in introducing new concepts. We present PersonaMagic, a stage-regulated generative technique designed for high-fidelity face customization. Using a simple MLP network, our method learns a series of embeddings within a specific timestep interval to capture face concepts. Additionally, we develop a Tandem Equilibrium mechanism that adjusts self-attention responses in the text encoder, balancing text description and identity preservation, improving both areas. Extensive experiments confirm the superiority of PersonaMagic over state-of-the-art methods in both qualitative and quantitative evaluations. Moreover, its robustness and flexibility are validated in non-facial domains, and it can also serve as a valuable plug-in for enhancing the performance of pretrained personalization models.

Event cameras, known for their low latency and high dynamic range, show great potential in pedestrian detection applications. However, while recent research has primarily focused on improving detection accuracy, the robustness of event-based visual models against physical adversarial attacks has received limited attention. For example, adversarial physical objects, such as specific clothing patterns or accessories, can exploit inherent vulnerabilities in these systems, leading to misdetections or misclassifications. This study is the first to explore physical adversarial attacks on event-driven pedestrian detectors, specifically investigating whether certain clothing patterns worn by pedestrians can cause these detectors to fail, effectively rendering them unable to detect the person. To address this, we developed an end-to-end adversarial framework in the digital domain, framing the design of adversarial clothing textures as a 2D texture optimization problem. By crafting an effective adversarial loss function, the framework iteratively generates optimal textures through backpropagation. Our results demonstrate that the textures identified in the digital domain possess strong adversarial properties. Furthermore, we translated these digitally optimized textures into physical clothing and tested them in real-world scenarios, successfully demonstrating that the designed textures significantly degrade the performance of event-based pedestrian detection models. This work highlights the vulnerability of such models to physical adversarial attacks.

Multimodal semantic segmentation is a critical challenge in computer vision, with early methods suffering from high computational costs and limited transferability due to full fine-tuning of RGB-based pre-trained parameters. Recent studies, while leveraging additional modalities as supplementary prompts to RGB, still predominantly rely on RGB, which restricts the full potential of other modalities. To address these issues, we propose a novel symmetric parameter-efficient fine-tuning framework for multimodal segmentation, featuring with a modality-aware prompting and adaptation scheme, to simultaneously adapt the capabilities of a powerful pre-trained model to both RGB and X modalities. Furthermore, prevalent approaches use the global cross-modality correlations of attention mechanism for modality fusion, which inadvertently introduces noise across modalities. To mitigate this noise, we propose a dynamic sparse cross-modality fusion module to facilitate effective and efficient cross-modality fusion. To further strengthen the above two modules, we propose a training strategy that leverages accurately predicted dual-modality results to self-teach the single-modality outcomes. In comprehensive experiments, we demonstrate that our method outperforms previous state-of-the-art approaches across six multimodal segmentation scenarios with minimal computation cost.

Previous successful approaches to missing modality completion rely on carefully designed fusion techniques and extensive pre-training on complete data, which can limit their generalizability in out-of-domain (OOD) scenarios. In this study, we pose a new challenge: can we develop a missing modality completion model that is both resource-efficient and robust to OOD generalization? To address this, we present a training-free framework for missing modality completion that leverages large multimodal model (LMM). Our approach, termed the "Knowledge Bridger", is modality-agnostic and integrates generation and ranking of missing modalities. By defining domain-specific priors, our method automatically extracts structured information from available modalities to construct knowledge graphs. These extracted graphs connect the missing modality generation and ranking modules through the LMM, resulting in high-quality imputations of missing modalities. Experimental results across both general and medical domains show that our approach consistently outperforms competing methods, including in OOD generalization. Additionally, our knowledge-driven generation and ranking techniques demonstrate superiority over variants that directly employ LMMs for generation and ranking, offering insights that may be valuable for applications in other domains.

Multimodal unsupervised domain adaptation leverages un-labeled data in the target domain to enhance multimodal systems continuously. While current state-of-the-art methods encourage interaction between sub-models of different modalities through pseudo-labeling and feature-level exchange, varying sample quality across modalities can lead to the propagation of inaccurate information, resulting in error accumulation. To address this, we propose Modal-Affinity Multimodal Domain Adaptation (MODfinity), a method that dynamically manages multimodal information flow through fine-grained control over teacher model selection, guiding information intertwining at both feature and label levels. By treating labels as an independent modality, MODfinity enables balanced performance assessment across modalities, employing a novel modal-affinity measurement to evaluate information quality. Additionally, we introduce a modal-affinity distillation technique to control sample-level information exchange, ensuring reliable multimodal interaction based on affinity evaluations within the feature space. Extensive experiments on three multimodal datasets demonstrate that our framework consistently outperforms state-of-the-art methods, particularly in high-noise environments.

Neural Radiance Field (NeRF) and 3D Gaussian Splatting (3DGS) have noticeably advanced photo-realistic novel view synthesis using images from densely spaced camera viewpoints. However, these methods struggle in few-shot scenarios due to limited supervision. In this paper, we present NexusGS, a 3DGS-based approach that enhances novel view synthesis from sparse-view images by directly embedding depth information into point clouds, without relying on complex manual regularizations. Exploiting the inherent epipolar geometry of 3DGS, our method introduces a novel point cloud densification strategy that initializes 3DGS with a dense point cloud, reducing randomness in point placement while preventing over-smoothing and overfitting. Specifically, NexusGS comprises three key steps: Epipolar Depth Nexus, Flow-Resilient Depth Blending, and Flow-Filtered Depth Pruning. These steps leverage optical flow and camera poses to compute accurate depth maps, while mitigating the inaccuracies often associated with optical flow. By incorporating epipolar depth priors, NexusGS ensures reliable dense point cloud coverage and supports stable 3DGS training under sparse-view conditions. Experiments demonstrate that NexusGS significantly enhances depth accuracy and rendering quality, surpassing state-of-the-art methods by a considerable margin. Furthermore, we validate the superiority of our generated point clouds by substantially boosting the performance of competing methods. Project page: https://usmizuki.github.io/NexusGS/.

In recent years, significant progress has been made in prototype-based learning methods for few-shot semantic segmentation. However, prototype features originating from the support images are interfered with by intra-class diversity and thus cannot be aligned with the query foreground, resulting in poor segmentation accuracy. Therefore, we propose a novel self-support prototype-aware (SSPA) network to obtain highly confident query foreground pixel points and their corresponding query features. We design Cycle Consistency Collection module and Self-Support Collection module to address the interference of invalid support prototypes. Experimental results demonstrate that our SSPA significantly improves the quality of prototypes and achieves state-of-the-art segmentation results on multiple datasets. In particular, SSPA achieves mIoU scores of 69.7% and 76.4% for 1-shot and 5-shot segmentation, respectively, on PASCAL-5i.

The rapid growth of 3D digital content necessitates expandable recognition systems for open-world scenarios. However, existing 3D class-incremental learning methods struggle under extreme data scarcity due to geometric misalignment and texture bias. While recent approaches integrate 3D data with 2D foundation models (e.g., CLIP), they suffer from semantic blurring caused by texture-biased projections and indiscriminate fusion of geometric-textural cues, leading to unstable decision prototypes and catastrophic forgetting. To address these issues, we propose Cross-Modal Geometric Rectification (CMGR), a framework that enhances 3D geometric fidelity by leveraging CLIP's hierarchical spatial semantics. Specifically, we introduce a Structure-Aware Geometric Rectification module that hierarchically aligns 3D part structures with CLIP's intermediate spatial priors through attention-driven geometric fusion. Additionally, a Texture Amplification Module synthesizes minimal yet discriminative textures to suppress noise and reinforce cross-modal consistency. To further stabilize incremental prototypes, we employ a Base-Novel Discriminator that isolates geometric variations. Extensive experiments demonstrate that our method significantly improves 3D few-shot class-incremental learning, achieving superior geometric coherence and robustness to texture bias across cross-domain and within-domain settings.

Image-based Virtual Try-On (VTON) techniques rely on either supervised in-shop approaches, which ensure high fidelity but struggle with cross-domain generalization, or unsupervised in-the-wild methods, which improve adaptability but remain constrained by data biases and limited universality. A unified, training-free solution that works across both scenarios remains an open challenge. We propose OmniVTON, the first training-free universal VTON framework that decouples garment and pose conditioning to achieve both texture fidelity and pose consistency across diverse settings. To preserve garment details, we introduce a garment prior generation mechanism that aligns clothing with the body, followed by continuous boundary stitching technique to achieve fine-grained texture retention. For precise pose alignment, we utilize DDIM inversion to capture structural cues while suppressing texture interference, ensuring accurate body alignment independent of the original image textures. By disentangling garment and pose constraints, OmniVTON eliminates the bias inherent in diffusion models when handling multiple conditions simultaneously. Experimental results demonstrate that OmniVTON achieves superior performance across diverse datasets, garment types, and application scenarios. Notably, it is the first framework capable of multi-human VTON, enabling realistic garment transfer across multiple individuals in a single scene. Code is available at https://github.com/Jerome-Young/OmniVTON

Generating sketches guided by reference styles requires precise transfer of stroke attributes, such as line thickness, deformation, and texture sparsity, while preserving semantic structure and content fidelity. To this end, we propose Stroke2Sketch, a novel training-free framework that introduces cross-image stroke attention, a mechanism embedded within self-attention layers to establish fine-grained semantic correspondences and enable accurate stroke attribute transfer. This allows our method to adaptively integrate reference stroke characteristics into content images while maintaining structural integrity. Additionally, we develop adaptive contrast enhancement and semantic-focused attention to reinforce content preservation and foreground emphasis. Stroke2Sketch effectively synthesizes stylistically faithful sketches that closely resemble handcrafted results, outperforming existing methods in expressive stroke control and semantic coherence. Codes are available at https://github.com/rane7/Stroke2Sketch.

Text-guided image and 3D editing have advanced with diffusion-based models, yet methods like Delta Denoising Score often struggle with stability, spatial control, and editing strength. These limitations stem from reliance on complex auxiliary structures, which introduce conflicting optimization signals and restrict precise, localized edits. We introduce Stable Score Distillation (SSD), a streamlined framework that enhances stability and alignment in the editing process by anchoring a single classifier to the source prompt. Specifically, SSD utilizes Classifier-Free Guidance (CFG) equation to achieves cross-prompt alignment, and introduces a constant term null-text branch to stabilize the optimization process. This approach preserves the original content's structure and ensures that editing trajectories are closely aligned with the source prompt, enabling smooth, prompt-specific modifications while maintaining coherence in surrounding regions. Additionally, SSD incorporates a prompt enhancement branch to boost editing strength, particularly for style transformations. Our method achieves state-of-the-art results in 2D and 3D editing tasks, including NeRF and text-driven style edits, with faster convergence and reduced complexity, providing a robust and efficient solution for text-guided editing.

Current text-to-3D generation methods based on score distillation often suffer from geometric inconsistencies, leading to repeated patterns across different poses of 3D assets. This issue, known as the Multi-Face Janus problem, arises because existing methods struggle to maintain consistency across varying poses and are biased toward a canonical pose. While recent work has improved pose control and approximation, these efforts are still limited by this inherent bias, which skews the guidance during generation. To address this, we propose a solution called RecDreamer, which reshapes the underlying data distribution to achieve more consistent pose representation. The core idea behind our method is to rectify the prior distribution, ensuring that pose variation is uniformly distributed rather than biased toward a canonical form. By modifying the prescribed distribution through an auxiliary function, we can reconstruct the density of the distribution to ensure compliance with specific marginal constraints. In particular, we ensure that the marginal distribution of poses follows a uniform distribution, thereby eliminating the biases introduced by the prior knowledge. We incorporate this rectified data distribution into existing score distillation algorithms, a process we refer to as uniform score distillation. To efficiently compute the posterior distribution required for the auxiliary function, RecDreamer introduces a training-free classifier that estimates pose categories in a plug-and-play manner. Additionally, we utilize various approximation techniques for noisy states, significantly improving system performance. Our experimental results demonstrate that RecDreamer effectively mitigates the Multi-Face Janus problem, leading to more consistent 3D asset generation across different poses.

Images are often obstructed by various obstacles due to capture limitations, hindering the observation of objects of interest. Most existing methods address occlusions from specific elements like fences or raindrops, but are constrained by the wide range of real-world obstructions, making comprehensive data collection impractical. To overcome these challenges, we propose Instruct2See, a novel zero-shot framework capable of handling both seen and unseen obstacles. The core idea of our approach is to unify obstruction removal by treating it as a soft-hard mask restoration problem, where any obstruction can be represented using multi-modal prompts, such as visual semantics and textual instructions, processed through a cross-attention unit to enhance contextual understanding and improve mode control. Additionally, a tunable mask adapter allows for dynamic soft masking, enabling real-time adjustment of inaccurate masks. Extensive experiments on both in-distribution and out-of-distribution obstacles show that Instruct2See consistently achieves strong performance and generalization in obstruction removal, regardless of whether the obstacles were present during the training phase. Code and dataset are available at https://jhscut.github.io/Instruct2See.

We introduce the task of Audible Action Temporal Localization, which aims to identify the spatio-temporal coordinates of audible movements. Unlike conventional tasks such as action recognition and temporal action localization, which broadly analyze video content, our task focuses on the distinct kinematic dynamics of audible actions. It is based on the premise that key actions are driven by inflectional movements; for example, collisions that produce sound often involve abrupt changes in motion. To capture this, we propose $TA^{2}Net$, a novel architecture that estimates inflectional flow using the second derivative of motion to determine collision timings without relying on audio input. $TA^{2}Net$ also integrates a self-supervised spatial localization strategy during training, combining contrastive learning with spatial analysis. This dual design improves temporal localization accuracy and simultaneously identifies sound sources within video frames. To support this task, we introduce a new benchmark dataset, $Audible623$, derived from Kinetics and UCF101 by removing non-essential vocalization subsets. Extensive experiments confirm the effectiveness of our approach on $Audible623$ and show strong generalizability to other domains, such as repetitive counting and sound source localization. Code and dataset are available at https://github.com/WenlongWan/Audible623.

Few-shot learning (FSL) addresses the challenge of classifying novel classes with limited training samples. While some methods leverage semantic knowledge from smaller-scale models to mitigate data scarcity, these approaches often introduce noise and bias due to the data’s inherent simplicity. In this paper, we propose a novel framework, Synergistic Knowledge Transfer (SynTrans), which effectively transfers diverse and complementary knowledge from large multimodal models to empower the off-the-shelf few-shot learner. Specifically, SynTrans employs CLIP as a robust teacher and uses a few-shot vision encoder as a weak student, distilling semantic-aligned visual knowledge via an unsupervised proxy task. Subsequently, a training-free synergistic knowledge mining module facilitates collaboration among large multimodal models to extract high-quality semantic knowledge. Building upon this, a visual-semantic bridging module enables bi-directional knowledge transfer between visual and semantic spaces, transforming explicit visual and implicit semantic knowledge into category-specific classifier weights. Finally, SynTrans introduces a visual weight generator and a semantic weight reconstructor to adaptively construct optimal multimodal FSL classifiers. Experimental results on four FSL datasets demonstrate that SynTrans, even when paired with a simple few-shot vision encoder, significantly outperforms current state-of-the-art methods.

Maritime object detection is essential for navigation safety, surveillance, and autonomous operations, yet constrained by two key challenges: the scarcity of annotated maritime data and poor generalization across various maritime attributes (e.g., object category, viewpoint, location, and imaging environment). To address these challenges, we propose Neptune-X, a data-centric generative-selection framework that enhances training effectiveness by leveraging synthetic data generation with task-aware sample selection. From the generation perspective, we develop X-to-Maritime, a multi-modality-conditioned generative model that synthesizes diverse and realistic maritime scenes. A key component is the Bidirectional Object-Water Attention module, which captures boundary interactions between objects and their aquatic surroundings to improve visual fidelity. To further improve downstream tasking performance, we propose Attribute-correlated Active Sampling, which dynamically selects synthetic samples based on their task relevance. To support robust benchmarking, we construct the Maritime Generation Dataset, the first dataset tailored for generative maritime learning, encompassing a wide range of semantic conditions. Extensive experiments demonstrate that our approach sets a new benchmark in maritime scene synthesis, significantly improving detection accuracy, particularly in challenging and previously underrepresented settings. The code is available at https://github.com/gy65896/Neptune-X.

3D Visual Grounding (3DVG) faces persistent challenges due to coarse scene-level observations and logically inconsistent annotations, which introduce ambiguities that compromise data quality and hinder effective model supervision. To address these challenges, we introduce Refer-Judge, a novel framework that harnesses the reasoning capabilities of Multimodal Large Language Models (MLLMs) to identify and mitigate toxic data. At the core of Refer-Judge is a Jury-and-Judge Chain-of-Thought paradigm, inspired by the deliberative process of the judicial system. This framework targets the root causes of annotation noise: jurors collaboratively assess 3DVG samples from diverse perspectives, providing structured, multi-faceted evaluations. Judges then consolidate these insights using a Corroborative Refinement strategy, which adaptively reorganizes information to correct ambiguities arising from biased or incomplete observations. Through this two-stage deliberation, Refer-Judge significantly enhances the reliability of data judgments. Extensive experiments demonstrate that our framework not only achieves human-level discrimination at the scene level but also improves the performance of baseline algorithms via data purification. Code is available at https://github.com/Hermione-HKX/Refer_Judge.

The explosive growth of generative video models has amplified the demand for reliable copyright preservation of AI-generated content. Despite its popularity in image synthesis, invisible generative watermarking remains largely underexplored in video generation. To address this gap, we propose Safe-Sora, the first framework to embed graphical watermarks directly into the video generation process. Motivated by the observation that watermarking performance is closely tied to the visual similarity between the watermark and cover content, we introduce a hierarchical coarse-to-fine adaptive matching mechanism. Specifically, the watermark image is divided into patches, each assigned to the most visually similar video frame, and further localized to the optimal spatial region for seamless embedding. To enable spatiotemporal fusion of watermark patches across video frames, we develop a 3D wavelet transform-enhanced Mamba architecture with a novel scanning strategy, effectively modeling long-range dependencies during watermark embedding and retrieval. To the best of our knowledge, this is the first attempt to apply state space models to watermarking, opening new avenues for efficient and robust watermark protection. Extensive experiments demonstrate that Safe-Sora achieves state-of-the- art performance in terms of video quality, watermark fidelity, and robustness, which is largely attributed to our proposals. Code and additional supporting materials are provided in the supplementary.

The advancement of diffusion models has enhanced the realism of AI-generated content but also raised concerns about misuse, necessitating robust copyright protection and tampering localization. Although recent methods have made progress toward unified solutions, their reliance on post hoc processing introduces considerable application inconvenience and compromises forensic reliability. We propose StableGuard, a novel framework that seamlessly integrates a binary watermark into the diffusion generation process, ensuring copyright protection and tampering localization in Latent Diffusion Models through an end-to-end design. We develop a Multiplexing Watermark VAE (MPW-VAE) by equipping a pretrained Variational Autoencoder (VAE) with a lightweight latent residual-based adapter, enabling the generation of paired watermarked and watermark-free images. These pairs, fused via random masks, create a diverse dataset for training a tampering-agnostic forensic network. To further enhance forensic synergy, we introduce a Mixture-of-Experts Guided Forensic Network (MoE-GFN) that dynamically integrates holistic watermark patterns, local tampering traces, and frequency-domain cues for precise watermark verification and tampered region detection. The MPW-VAE and MoE-GFN are jointly optimized in a self-supervised, end-to-end manner, fostering a reciprocal training between watermark embedding and forensic accuracy. Extensive experiments demonstrate that StableGuard consistently outperforms state-of-the-art methods in image fidelity, watermark verification, and tampering localization.

3D anomaly detection in point-cloud data is critical for industrial quality control, aiming to identify structural defects with high reliability. However, current memory bank-based methods often suffer from inconsistent feature transformations and limited discriminative capacity, particularly in capturing local geometric details and achieving rotation invariance. These limitations become more pronounced when registration fails, leading to unreliable detection results. We argue that point-cloud registration plays an essential role not only in aligning geometric structures but also in guiding feature extraction toward rotation-invariant and locally discriminative representations. To this end, we propose a registration-induced, rotation-invariant feature extraction framework that integrates the objectives of point-cloud registration and memory-based anomaly detection. Our key insight is that both tasks rely on modeling local geometric structures and leveraging feature similarity across samples. By embedding feature extraction into the registration learning process, our framework jointly optimizes alignment and representation learning. This integration enables the network to acquire features that are both robust to rotations and highly effective for anomaly detection. Extensive experiments on the Anomaly-ShapeNet and Real3D-AD datasets demonstrate that our method consistently outperforms existing approaches in effectiveness and generalizability.

This study presents anchor-regularized generative adversarial network (GAN) priors to delicately explore the inherent knowledge of a pretrained generative model. Previous research leveraged the latent space of a pretrained GAN model to provide a variety of image-editing operations. However, the semantically meaningful regions within latent space are distinctly bounded; therefore, the manipulation of the latent code can easily land out of the domain. To address this problem, we introduce an anchoring mechanism that enables novel and robust image editing. The key insights driving the method are that latent space is structurally organized, and that natural coherence allows semantically correlated latent code to be located in the areas surrounding a meaningful anchor. By using different input anchors, the proposed method forms the basis for a variety of robust and flexible editing operations, including misaligned domain translation, interactive editing, and few-shot interpretable direction exploration. Extensive experiments demonstrated the superior performance of the proposed method compared with state-of-the-art editing methods.

In this paper, we propose a novel translation model, UniTranslator, for transforming representations between visually distinct domains under conditions of limited training data and significant visual differences. The main idea behind our approach is leveraging the domain-neutral capabilities of CLIP as a bridging mechanism, while utilizing a separate module to extract abstract, domain-agnostic semantics from the embeddings of both the source and target realms. Fusing these abstract semantics with target-specific semantics results in a transformed embedding within the CLIP space. To bridge the gap between the disparate worlds of CLIP and StyleGAN, we introduce a new non-linear mapper, the CLIP2P mapper. Utilizing CLIP embeddings, this module is tailored to approximate the latent distribution in the StyleGAN's latent space, effectively acting as a connector between these two spaces. The proposed UniTranslator is versatile and capable of performing various tasks, including style mixing, stylization, and translations, even in visually challenging scenarios across different visual domains. Notably, UniTranslator generates high-quality translations that showcase domain relevance, diversity, and improved image quality. UniTranslator surpasses the performance of existing general-purpose models and performs well against specialized models in representative tasks.

The vulnerability of 3D point cloud analysis to unpredictable rotations poses an open yet challenging problem: orientation-aware 3D domain generalization. Cross-domain robustness and adaptability of 3D representations are crucial but not easily achieved through rotation augmentation. Motivated by the inherent advantages of intricate orientations in enhancing generalizability, we propose an innovative rotation-adaptive domain generalization framework for 3D point cloud analysis. Our approach aims to alleviate orientational shifts by leveraging intricate samples in an iterative learning process. Specifically, we identify the most challenging rotation for each point cloud and construct an intricate orientation set by optimizing intricate orientations. Subsequently, we employ an orientation-aware contrastive learning framework that incorporates an orientation consistency loss and a margin separation loss, enabling effective learning of categorically discriminative and generalizable features with rotation consistency. Extensive experiments and ablations conducted on 3D cross-domain benchmarks firmly establish the state-of-the-art performance of our proposed approach in the context of orientation-aware 3D domain generalization.

Interactive 3D segmentation in radiance fields is crucial for advanced 3D scene understanding and manipulation. However, existing methods often struggle to achieve both volumetric completeness and segmentation accuracy, primarily because they fail to consider the critical links between 2D prompt-based segmentations across multiple views. Motivated by this gap, we introduce Gaussian Prompter, a novel approach specifically designed for 3D Gaussian Splatting. The core idea behind Gaussian Prompter is to seamlessly integrate a Gaussian-centric segmentation paradigm by effectively linking various 2D prompts from multi-view segmentations to ensure consistent 3D segmentation. To realize this, we employ two tailored approaches: GaussBlend and PinPrompt. GaussBlend aggregates multi-view 2D segmentation masks into a cohesive 3D segmentation, ensuring both accuracy and completeness. PinPrompt leverages high-confidence prompts from adjacent views to enhance segmentation precision further. Additionally, to address the lack of complex datasets in 3D segmentation, we introduce the SegMip-360 dataset, which includes over 350 precisely annotated masks across seven scenes. Extensive experiments demonstrate that the Gaussian Prompter significantly outperforms state-of-the-art methods in both segmentation accuracy and completeness. Our code and video demonstrations can be found at our repository and project page.

RGB-Thermal Salient Object Detection (RGB-T SOD) aims to pinpoint prominent objects within aligned pairs of visible and thermal infrared images. A key challenge lies in bridging the inherent disparities between RGB and Thermal modalities for effective saliency map prediction. Traditional encoder-decoder architectures, while designed for cross-modality feature interactions, may not have adequately considered the robustness against noise originating from defective modalities, thereby leading to suboptimal performance in complex scenarios. Inspired by hierarchical human visual systems, we propose the ConTriNet, a robust Confluent Triple-Flow Network employing a "Divide-and-Conquer" strategy. This framework utilizes a unified encoder with specialized decoders, each addressing different subtasks of exploring modality-specific and modality-complementary information for RGB-T SOD, thereby enhancing the final saliency map prediction. Specifically, ConTriNet comprises three flows: two modality-specific flows explore cues from RGB and Thermal modalities, and a third modality-complementary flow integrates cues from both modalities. ConTriNet presents several notable advantages. It incorporates a Modality-induced Feature Modulator (MFM) in the modality-shared union encoder to minimize inter-modality discrepancies and mitigate the impact of defective samples. Additionally, a foundational Residual Atrous Spatial Pyramid Module (RASPM) in the separated flows enlarges the receptive field, allowing for the capture of multi-scale contextual information. Furthermore, a Modality-aware Dynamic Aggregation Module (MDAM) in the modality-complementary flow dynamically aggregates saliency-related cues from both modality-specific flows. Leveraging the proposed parallel triple-flow framework, we further refine saliency maps derived from different flows through a flow-cooperative fusion strategy, yielding a high-quality, full-resolution saliency map for the final prediction. To evaluate the robustness and stability of our approach, we collect a comprehensive RGB-T SOD benchmark, VT-IMAG, covering various real-world challenging scenarios. Extensive experiments on public benchmarks and our VT-IMAG dataset demonstrate that ConTriNet consistently outperforms state-of-the-art competitors in both common and challenging scenarios, even when dealing with incomplete modality data. The code and VT-IMAG will be available at: https://cser-tang-hao.github.io/contrinet.html.

We introduce a new task, Open-set Mixed Domain Adaptation (OSMDA), which considers the potential mixture of multiple distributions in the target domains, thereby better simulating real-world scenarios. To tackle the semantic ambiguity arising from multiple domains, our key idea is that the linguistic representation can serve as a universal descriptor for samples of the same category across various domains. We thus propose a more practical framework for cross-domain recognition via visual-linguistic guidance. On the other hand, the presence of multiple domains also poses a new challenge in classifying both known and unknown categories. To combat this issue, we further introduce a visual-linguistic focal evolving approach to gradually enhance the classification ability of a known/unknown binary classifier from two aspects. Specifically, we start with identifying highly confident focal samples to expand the pool of known samples by incorporating those from different domains. Then, we amplify the feature discrepancy between known and unknown samples through dynamic entropy evolving via an adaptive entropies min/max game, enabling us to accurately identify possible unknown samples in a gradual manner. Extensive experiments demonstrate our method’s superiority against the state-of-the-arts in both open-set and open-set mixed domain adaptation.

Recent advances have heightened the interest in the adversarial transferability of Vision-Language Pre-training (VLP) models. However, most existing strategies constrained by two persistent limitations: suboptimal utilization of cross-modal interactive information, and inherent discrepancies across hierarchical textual representation. To address these challenges, we propose the Modality-Specific Interactive Attack (MSI-Attack), a novel approach that integrates semantic-level image perturbations with embedding-level text perturbations, all while maintaining minimal inter-modal constraints. In our image attack methodology, we introduce Multi-modal Integrated Gradients (MIG) to guide perturbations toward the core semantics of images, enriched by their associated deeply text information. This technique enhances transferability by capturing consistent features across various models, thereby effectively misleading similar-model perception areas. Additionally, we employ a momentum iteration strategy in conjunction with MIG, which amalgamates current and historical gradients to expedite the perturbation updates. For text attacks, we streamline the perturbation process by operating exclusively at the embedding level. This reduces semantic gaps across hierarchical structures and significantly enhances the generalizability of adversarial text. Moreover, we delve deeper into how semantic perturbations with varying degrees of similarity affect the overall attack effectiveness. Our experimental results on image-text retrieval tasks using the multi-modal datasets Flickr30K and MSCOCO underscore the efficacy of MSI-Attack. Our method achieves superior performance, setting a new state-of-the-art benchmark, all without the need for additional mechanisms.

Unsupervised domain adaptation enables the transfer of knowledge from a labeled source domain to an unlabeled target domain, and its application in crowd counting is gaining momentum. Current methods typically align distributions across domains to address inter-domain disparities at a global level. However, these methods often struggle with significant intra-domain gaps caused by domain-agnostic factors such as density, surveillance angles, and scale, leading to inaccurate alignment and unnecessary computational burdens, especially in large-scale training scenarios. To address these challenges, we propose the Multi-Granularity Optimal Transport (MGOT) distribution alignment framework, which aligns domain-agnostic factors across domains at different granularities. The motivation behind multi-granularity is to capture fine-grained domain-agnostic variations within domains. Our method proceeds in three phases: first, clustering coarse-grained features based on intra-domain similarity; second, aligning the granular clusters using an optimal transport framework and constructing a mapping from cluster centers to finer patch levels between domains; and third, re-weighting the aligned distribution for model refinement in domain adaptation. Extensive experiments across twelve cross-domain benchmarks show that our method outperforms existing state-of-the-art methods in adaptive crowd counting. The code will be available at https://github.com/HopooLinZ/MGOT.

Enabling efficient and accurate deep neural network (DNN) inference on microcontrollers is challenging due to their constrained on-chip resources. Existing approaches mainly focus on compressing larger models, often compromising model accuracy as a trade-off. In this paper, we rethink the problem from the inverse perspective by directly constructing small/weak models, then enhancing their accuracy. Thus, we propose DiTMoS, a novel DNN training and inference framework featuring aselector-classifiersarchitecture, where the selector routes each input sample to the appropriate classifier for classification. DiTMoS is built on a key insight: a combination of weak models can exhibit high diversity and the union of them can significantly raise the upper bound of overall accuracy. To approach the upper bound, DiTMoS introduces three strategies including diverse training data splitting to enhance the classifiers' diversity, adversarial selector-classifiers training to ensure synergistic interactions thereby maximizing their complementarity, and heterogeneous feature aggregation to improve the capacity of classifiers. We further design a network slicing technique to eliminate the extra memory consumption incurred by feature aggregation. We deploy DiTMoS on the Nucleo STM32F767ZI board and evaluate its performance across three time-series datasets for human activity recognition, keyword spotting, and emotion recognition tasks. The experimental results show that: (a) DiTMoS improves accuracy by up to 13.4% compared to the best baseline; (b) network slicing successfully eliminates the memory overhead introduced by feature aggregation, with only a minimal increase in latency. The code of DiTMoS is released athttps://github.com/TheMaXiao/DiTMoS

Monocular depth prediction has received significant attention in recent years. However, the impact of illumination variations, which can shift scenes to unseen domains, has often been overlooked. To address this, we introduce the first indoor scene dataset featuring RGB-D images captured under multiple illumination conditions, allowing for a comprehensive exploration of indoor depth prediction. Additionally, we propose a novel method, MI-Transformer, which leverages global illumination understanding through large receptive fields to capture depth-attention contexts. This enables our network to overcome local window limitations and effectively mitigate the influence of changing illumination conditions. To evaluate the performance and robustness, we conduct extensive qualitative and quantitative analyses on both the proposed dataset and existing benchmarks, comparing our method with state-of-the-art approaches. The experimental results demonstrate the superiority of our method across various metrics, making it the first solution to achieve robust monocular depth estimation under diverse illumination conditions. We provide the codes, pre-trained models, and dataset openly accessible athttps://github.com/ViktorLiang/midepth.

Crowd counting has drawn increasing attention across various fields. However, existing crowd counting tasks primarily focus on estimating the overall population, ignoring the behavioral and semantic information of different social groups within the crowd. In this paper, we aim to address a newly proposed research problem, namely fine-grained crowd counting, which involves identifying different categories of individuals and accurately counting them in static images. In order to fully leverage the categorical information in static crowd images, we propose a two-tier salient feature propagation module designed to sequentially extract semantic information from both the crowd and its surrounding environment. Additionally, we introduce a category difference loss to refine the feature representation by highlighting the differences between various crowd categories. Moreover, our proposed framework can adapt to a novel problem setup called few-example fine-grained crowd counting. This setup, unlike the original fine-grained crowd counting, requires only a few exemplar point annotations instead of dense annotations from predefined categories, making it applicable in a wider range of scenarios. The baseline model for this task can be established by substituting the loss function in our proposed model with a novel hybrid loss function that integrates point-oriented cross-entropy loss and category contrastive loss. Through comprehensive experiments, we present results in both the formulation and application of fine-grained crowd counting.

Mild cognitive impairment (MCI) represents an early stage of Alzheimer's disease (AD), characterized by subtle clinical symptoms that pose challenges for accurate diagnosis. The quest for the identification of MCI individuals has highlighted the importance of comprehending the underlying mechanisms of disease causation. Integrated analysis of brain imaging and genomics offers a promising avenue for predicting MCI risk before clinical symptom onset. However, most existing methods face challenges in: 1) mining the brain network-specific topological structure and addressing the single nucleotide polymorphisms (SNPs)-related noise contamination and 2) extracting the discriminative properties of brain imaging genomics, resulting in limited accuracy for MCI diagnosis. To this end, a modality-aware discriminative fusion network (MA-DFN) is proposed to integrate the complementary information from brain imaging genomics to diagnose MCI. Specifically, we first design two modality-specific feature extraction modules: the graph convolutional network with edge-augmented self-attention module (GCN-EASA) and the deep adversarial denoising autoencoder module (DAD-AE), to capture the topological structure of brain networks and the intrinsic distribution of SNPs. Subsequently, a discriminative-enhanced fusion network with correlation regularization module (DFN-CorrReg) is employed to enhance inter-modal consistency and between-class discrimination in brain imaging and genomics. Compared to other state-of-the-art approaches, MA-DFN not only exhibits superior performance in stratifying cognitive normal (CN) and MCI individuals but also identifies disease-related brain regions and risk SNPs locus, which hold potential as putative biomarkers for MCI diagnosis.

Although pre-trained large-scale generative models StyleGAN series have proven to be effective in various editing and translation tasks, they are limited to pre-defined fixed aspect ratio. To overcome this limitation, we propose StyleGAN-$\infty$∞, a model that enables pre-trained StyleGAN to perform arbitrary-ratio conditional synthesis. Our key insight is to distill the expressive StyleGAN features into a StyleBook, such that an arbitrary-ratio condition can be translated to other forms by properly assembling pre-defined StyleBook vectors. To learn and leverage the StyleBook, we employ a network with three distinct stages, each corresponding to StyleBook extraction, StyleBook correspondence learning, and arbitrary-ratio synthesis. Extensive experiments on various conditional synthesis tasks, like super-resolution, sketch synthesis, and semantic synthesis, demonstrate superior performances over state-of-the-art image-to-image translation methods. Moreover, our model can easily generate megapixel images in diverse modalities by taking advantage of different pre-trained StyleGAN models.

Traditional human neural radiance fields often overlook crucial body semantics, resulting in ambiguous reconstructions, particularly in occluded regions. To address this problem, we propose the Super-Semantic Disentangled Neural Renderer (SSD-NeRF), which employs rich regional semantic priors to enhance human rendering accuracy. This approach initiates with a Visible-Invisible Semantic Propagation module, ensuring coherent semantic assignment to occluded parts based on visible body segments. Furthermore, a Region-Wise Texture Propagation module independently extends textures from visible to occluded areas within semantic regions, thereby avoiding irrelevant texture mixtures and preserving semantic consistency. Additionally, a view-aware curricular learning approach is integrated to bolster the model's robustness and output quality across different viewpoints. Extensive evaluations confirm that SSD-NeRF surpasses leading methods, particularly in generating quality and structurally semantic reconstructions of unseen or occluded views and poses.

Automatic 3D shape analysis is heavily influenced by the pose of input 3D models, as the continuous nature of pose space introduces complexities that usually exceed the encoding capacities of standard deep learning frameworks. To tackle this challenge, we present Upright-Net+, an enhancement of our previous model, Upright-Net, specifically developed for estimating upright orientation in 3D point clouds. Our approach is grounded in the design principle that "form ever follows function," treating the natural base of an object as a functional structure that stabilizes it in its typical pose, influenced by physical laws and geometric properties. We reformulate the continuous orientation problem into a discrete classification task, focusing on learning the points that constitute the natural base of a 3D model. The upright orientation is determined by aligning the normal orientation of this base towards the mass center. To mitigate over-smoothing in the global feature embeddings from stacked graph convolutional layers, we introduce a Global Positional Encoding Module using Relative Distance Histogram Statistics Embedding (GPE-RDHS), which reduces structural ambiguity and enhances orientation estimation. We also enhanced a weighted residual loss term to penalize false positive predictions, enhancing overall model performance. Our method demonstrates exceptional performance in upright orientation estimation and reveals that the learned orientation-aware features significantly benefit downstream tasks, particularly in classification.

Text-to-image generation models have significantly broadened the horizons of creative expression through the power of natural language. However, navigating these models to generate unique concepts, alter their appearance, or reimagine them in unfamiliar roles presents an intricate challenge. For instance, how can we exploit language-guided models to transpose an anime character into a different art style, or envision a beloved character in a radically different setting or role? This paper unveils a novel approach named DreamAnime, designed to provide this level of creative freedom. Using a minimal set of 2-3 images of a user-specified concept such as an anime character or an art style, we teach our model to encapsulate its essence through novel "words" in the embedding space of a pre-existing text-to-image model. Crucially, we disentangle the concepts of style and identity into two separate "words", thus providing the ability to manipulate them independently. These distinct "words" can then be pieced together into natural language sentences, promoting an intuitive and personalized creative process. Empirical results suggest that this disentanglement into separate word embeddings successfully captures a broad range of unique and complex concepts, with each word focusing on style or identity as appropriate. Comparisons with existing methods illustrate DreamAnime's superior capacity to accurately interpret and recreate the desired concepts across various applications and tasks.

Current sketch extraction methods either require extensive training or fail to capture a wide range of artistic styles, limiting their practical applicability and versatility. We introduce Mixture-of-Self-Attention (MixSA), a training-free sketch extraction method that leverages strong diffusion priors for enhanced sketch perception. At its core, MixSA employs a mixture-of-self-attention technique, which manipulates self-attention layers by substituting the keys and values with those from reference sketches. This allows for the seamless integration of brushstroke elements into initial outline images, offering precise control over texture density and enabling interpolation between styles to create novel, unseen styles. By aligning brushstroke styles with the texture and contours of colored images, particularly in late decoder layers handling local textures, MixSA addresses the common issue of color averaging by adjusting initial outlines. Evaluated with various perceptual metrics, MixSA demonstrates superior performance in sketch quality, flexibility, and applicability. This approach not only overcomes the limitations of existing methods but also empowers users to generate diverse, high-fidelity sketches that more accurately reflect a wide range of artistic expressions.

Throughout history, static paintings have captivated viewers within display frames, yet the possibility of making these masterpieces vividly interactive remains intriguing. This research paper introduces 3DArtmator, a novel approach that aims to represent artforms in a highly interpretable stylized space, enabling 3D-aware animatable reconstruction and editing. Our rationale is to transfer the interpretability and 3D controllability of the latent space in a 3D-aware GAN to a stylized sub-space of a customized GAN, revitalizing the original artforms. To this end, the proposed two-stage optimization framework of 3DArtmator begins with discovering an anchor in the original latent space that accurately mimics the pose and content of a given art painting. This anchor serves as a reliable indicator of the original latent space local structure, therefore sharing the same editable predefined expression vectors. In the second stage, we train a customized 3D-aware GAN specific to the input artform, while enforcing the preservation of the original latent local structure through a meticulous style-directional difference loss. This approach ensures the creation of a stylized sub-space that remains interpretable and retains 3D control. The effectiveness and versatility of 3DArtmator are validated through extensive experiments across a diverse range of art styles. With the ability to generate 3D reconstruction and editing for artforms while maintaining interpretability, 3DArtmator opens up new possibilities for artistic exploration and engagement.

Fine-grained visual classification (FGVC) involves categorizing fine subdivisions within a broader category, which poses challenges due to subtle inter-class discrepancies and large intra-class variations. However, prevailing approaches primarily focus on uni-modal visual concepts. Recent advancements in pre-trained vision-language models have demonstrated remarkable performance in various high-level vision tasks, yet the applicability of such models to FGVC tasks remains uncertain. In this paper, we aim to fully exploit the capabilities of cross-modal description to tackle FGVC tasks and propose a novel multimodal prompting solution, denoted as MP-FGVC, based on the contrastive language-image pertaining (CLIP) model. Our MP-FGVC comprises a multimodal prompts scheme and a multimodal adaptation scheme. The former includes Subcategory-specific Vision Prompt (SsVP) and Discrepancy-aware Text Prompt (DaTP), which explicitly highlights the subcategory-specific discrepancies from the perspectives of both vision and language. The latter aligns the vision and text prompting elements in a common semantic space, facilitating cross-modal collaborative reasoning through a Vision-Language Fusion Module (VLFM) for further improvement on FGVC. Moreover, we tailor a two-stage optimization strategy for MP-FGVC to fully leverage the pre-trained CLIP model and expedite efficient adaptation for FGVC. Extensive experiments conducted on four FGVC datasets demonstrate the effectiveness of our MP-FGVC.

Multi-view representation learning aims to derive robust representations that are both view-consistent and view-specific from diverse data sources. This paper presents an in-depth analysis of existing approaches in this domain, highlighting a commonly overlooked aspect: the redundancy between view-consistent and view-specific representations. To this end, we propose an innovative framework for multi-view representation learning, which incorporates a technique we term ‘distilled disentangling’. Our method introduces the concept of masked cross-view prediction, enabling the extraction of compact, high-quality view-consistent representations from various sources without incurring extra computational overhead. Additionally, we develop a distilled disentangling module that efficiently filters out consistency-related information from multi-view representations, resulting in purer view-specific representations. This approach significantly reduces redundancy between view-consistent and view-specific representations, enhancing the overall efficiency of the learning process. Our empirical evaluations reveal that higher mask ratios substantially improve the quality of view-consistent representations. Moreover, we find that reducing the dimensionality of view-consistent representations relative to that of view-specific representations further refines the quality of the combined representations. Our code is accessible at: https://github.com/Guanzhou-Ke/MRDD.

Point-based interactive editing serves as an essential tool to complement the controllability of existing generative mod-els. A concurrent work, DragD iffus ion, updates the diffusion latent map in response to user inputs, causing global latent map alterations. This results in imprecise preservation of the original content and unsuccessful editing due to gradient vanishing. In contrast, we present DragNoise, offering ro-bust and accelerated editing without retracing the latent map. The core rationale of DragNoise lies in utilizing the predicted noise output of each U-Net as a semantic editor. This approach is grounded in two critical observations: firstly, the bottleneck features of U-Net inherently possess semantically rich features ideal for interactive editing; secondly, high-level semantics, established early in the denoising process, show minimal variation in subsequent stages. Leveraging these insights, DragNoise edits diffusion semantics in a sin-gle denoising step and efficiently propagates these changes, ensuring stability and efficiency in diffusion editing. Compar-ative experiments reveal that DragNoise achieves superior control and semantic retention, reducing the optimization time by over 50% compared to DragDiffusion. Our codes are available at https://github.com/haofenglIDragNoise.

Existing methods for asymmetric image retrieval employ a rigid pairwise similarity constraint between the query network and the larger gallery network. However, these one-to-one constraint approaches often fail to maintain retrieval order consistency, especially when the query network has limited representational capacity. To overcome this problem, we introduce the Decoupled Differential Distillation (D3still) framework. This framework shifts from absolute one-to-one supervision to optimizing the relational differences in pairwise similarities produced by the query and gallery networks, thereby preserving a consistent retrieval order across both networks. Our method involves computing a pairwise similarity differential matrix within the gallery domain, which is then decomposed into three components: feature representation knowledge, inconsistent pairwise similarity differential knowledge, and consistent pairwise similarity differential knowledge. This strategic decomposition aligns the retrieval ranking of the query network with the gallery network effectively. Extensive experiments on various bench-mark datasets reveal that D3still surpasses state-of-the-art methods in asymmetric image retrieval. Code is available at https://github.com/SCY-X/D3still.

We propose a voxel-based optimization framework, Re VoRF, for few-shot radiance fields that strategically ad-dress the unreliability in pseudo novel view synthesis. Our method pivots on the insight that relative depth relationships within neighboring regions are more reliable than the ab-solute color values in disoccluded areas. Consequently, we devise a bilateral geometric consistency loss that carefully navigates the trade-off between color fidelity and geometric accuracy in the context of depth consistency for uncertain regions. Moreover, we present a reliability-guided learning strategy to discern and utilize the variable quality across syn-thesized views, complemented by a reliability-aware voxel smoothing algorithm that smoothens the transition between reliable and unreliable data patches. Our approach allows for a more nuanced use of all available data, promoting en-hanced learning from regions previously considered unsuit-able for high-quality reconstruction. Extensive experiments across diverse datasets reveal that our approach attains significant gains in efficiency and accuracy, delivering ren-dering speeds of 3 FPS, 7 mins to train a 360° scene, and a 5% improvement in PSNR over existing few-shot methods. Code is available at https://github.com/HKCLynn/ReVoRF.

In this paper, we delve into a novel aspect of learning novel diffusion conditions with datasets an order of magnitude smaller. The rationale behind our approach is the elimination of textual constraints during the few-shot learning process. To that end, we implement two optimization strategies. The first, prompt-free conditional learning, utilizes a prompt-free encoder derived from a pre-trained Stable Diffusion model. This strategy is designed to adapt new conditions to the diffusion process by minimizing the textual-visual cor-relation, thereby ensuring a more precise alignment between the generated content and the specified conditions. The second strategy entails condition-specific negative rectification, which addresses the inconsistencies typically brought about by Classifier-free guidance in few-shot training con-texts. Our extensive experiments across a variety of condition modalities demonstrate the effectiveness and efficiency of our framework, yielding results comparable to those obtained with datasets a thousand times larger. Our codes are available at https://github.com/Yuyan9Yu/BeyondTextConstraint.

We introduce the Self-Exemplar Illumination Equalization Network, designed specifically for effective portrait shadow removal. The core idea of our method is that partially shadowed portraits can find ideal exemplars within their non-shadowed facial regions. Rather than directly fusing two distinct classes of facial features, our approach utilizes non-shadowed regions as an illumination indicator to equalize the shadowed regions, generating deshadowed results without boundary-merging artifacts. Our network comprises cascaded Self-Exemplar Illumination Equalization Blocks (SExmBlock), each containing two modules: a self-exemplar feature matching module and a feature-level illumination rectification module. The former identifies and applies internal illumination exemplars to shadowed areas, producing illumination-corrected features, while the latter adjusts shadow illumination by reapplying the illumination factors from these features to the input face. Applying this series of SExmBlocks to shadowed portraits incrementally eliminates shadows and preserves clear, accurate facial details. The effectiveness of our method is demonstrated through evaluations on two public shadow portrait datasets, where it surpasses existing state-of-the-art methods in both qualitative and quantitative assessments.

Video Visual Relation Detection (VidVRD) focuses on understanding how entities interact over time and space in videos, a key step for gaining deeper insights into video scenes beyond basic visual tasks. Traditional methods for VidVRD, challenged by its complexity, typically split the task into two parts: one for identifying what relation categories are present and another for determining their temporal boundaries. This split overlooks the inherent connection between these elements. Addressing the need to recognize entity pairs' spatiotemporal interactions across a range of durations, we propose VrdONE, a streamlined yet efficacious one-stage model. VrdONE combines the features of subjects and objects, turning predicate detection into 1D instance segmentation on their combined representations. This setup allows for both relation category identification and binary mask generation in one go, eliminating the need for extra steps like proposal generation or post-processing. VrdONE facilitates the interaction of features across various frames, adeptly capturing both short-lived and enduring relations. Additionally, we introduce the Subject-Object Synergy (SOS) module, enhancing how subjects and objects perceive each other before combining. VrdONE achieves state-of-the-art performances on the VidOR benchmark and ImageNet-VidVRD, showcasing its superior capability in discerning relations across different temporal scales. The code is available at https://github.com/lucaspk512/vrdone.

Drone-based crowd tracking faces difficulties in accurately identifying and monitoring objects from an aerial perspective, largely due to their small size and close proximity to each other, which complicates both localization and tracking. To address these challenges, we present the Density-aware Tracking (DenseTrack) framework. DenseTrack capitalizes on crowd counting to precisely determine object locations, blending visual and motion cues to improve the tracking of small-scale objects. It specifically addresses the problem of cross-frame motion to enhance tracking accuracy and dependability. DenseTrack employs crowd density estimates as anchors for exact object localization within video frames. These estimates are merged with motion and position information from the tracking network, with motion offsets serving as key tracking cues. Moreover, DenseTrack enhances the ability to distinguish small-scale objects using insights from the visual-language model, integrating appearance with motion cues. The framework utilizes the Hungarian algorithm to ensure the accurate matching of individuals across frames. Demonstrated on DroneCrowd dataset, our approach exhibits superior performance, confirming its effectiveness in scenarios captured by drones. Our code will be available at: https://github.com/Zebrabeast/DenseTrack.

Enabling efficient and accurate deep neural network (DNN) inference on microcontrollers is non-trivial due to the constrained on-chip resources. Current methodologies primarily focus on compressing larger models yet at the expense of model accuracy. In this paper, we rethink the problem from the inverse perspective by constructing small/weak models directly and improving their accuracy. Thus, we introduce DiTMoS, a novel DNN training and inference framework with a selector-classifiers architecture, where the selector routes each input sample to the appropriate classifier for classification. DiTMoS is grounded on a key insight: a composition of weak models can exhibit high diversity and the union of them can significantly boost the accuracy upper bound. To approach the upper bound, DiT-MoS introduces three strategies including diverse training data splitting to increase the classifiers' diversity, adversarial selector-classifiers training to ensure synergistic interactions thereby maximizing their complementarity, and heterogeneous feature aggregation to improve the capacity of classifiers. We further propose a network slicing technique to alleviate the extra memory overhead incurred by feature aggregation. We deploy DiTMoS on the Neucleo STM32F767ZI board and evaluate it based on three time-series datasets for human activity recognition, keywords spotting, and emotion recognition, respectively. The experiment results manifest that: (a) DiTMoS achieves up to 13.4% accuracy improvement compared to the best baseline; (b) network slicing almost completely eliminates the memory overhead incurred by feature aggregation with a marginal increase of latency. Code is released at https//github.com/TheMaXiao/DiTMoS

Crowd image is arguably one of the most laborious data to annotate. In this paper, we aim to reduce the massive demand for densely labeled crowd data, and propose a novel weakly-supervised setting, in which we leverage the binary ranking of two images with high-contrast crowd counts as training guidance. To enable training under this new setting, we convert the crowd count regression problem to a ranking potential prediction problem. In particular, we tailor a Siamese Ranking Network that predicts the potential scores of two images indicating the ordering of the counts. Hence, the ultimate goal is to assign appropriate potentials for all the crowd images to ensure their orderings obey the ranking labels. On the other hand, potentials reveal the relative crowd sizes but cannot yield an exact crowd count. We resolve this problem by introducing "anchors" during the inference stage. Concretely, anchors are a few images with count labels used for referencing the corresponding counts from potential scores by a simple linear mapping function. We conduct extensive experiments to study various combinations of supervision, and we show that our method outperforms existing weakly-supervised methods by a large margin without additional labeling effort. The code is available at https://github.com/pandaszzzzz/CCRanking.

HD map reconstruction is crucial for autonomous driving. LiDAR-based methods are limited due to expensive sensors and time-consuming computation. Camera-based methods usually need to perform road segmentation and view transformation separately, which often causes distortion and missing content. To push the limits of the technology, we present a novel framework that reconstructs a local map formed by road layout and vehicle occupancy in the bird's-eye view given a front-view monocular image only. We propose a front-to-top view projection (FTVP) module, which takes the constraint of cycle consistency between views into account and makes full use of their correlation to strengthen the view transformation and scene understanding. In addition, we apply multi-scale FTVP modules to propagate the rich spatial information of low-level features to mitigate spatial deviation of the predicted object location. Experiments on public benchmarks show that our method achieves various tasks on road layout estimation, vehicle occupancy estimation, and multi-class semantic estimation, at a performance level comparable to the state-of-the-arts, while maintaining superior efficiency.

3D neural rendering enables photo-realistic reconstruction of a specific scene by encoding discontinuous inputs into a neural representation. Despite the remarkable rendering results, the storage of network parameters is not transmission-friendly and not extendable to metaverse applications. In this paper, we propose an invertible neural rendering approach that enables generating an interactive 3D model from a single image (i.e., 3D Snapshot). Our idea is to distill a pre-trained neural rendering model (e.g., NeRF) into a visualizable image form that can then be easily inverted back to a neural network. To this end, we first present a neural image distillation method to optimize three neural planes for representing the original neural rendering model. However, this representation is noisy and visually meaningless. We thus propose a dynamic invertible neural network to embed this noisy representation into a plausible image representation of the scene. We demonstrate promising reconstruction quality quantitatively and qualitatively, by comparing to the original neural rendering model, as well as video-based invertible methods. On the other hand, our method can store dozens of NeRFs with a compact restoration network (5 MB), and embedding each 3D scene takes up only 160 KB of storage. More importantly, our approach is the first solution that allows embedding a neural rendering model into image representations, which enables applications like creating an interactive 3D model from a printed image in the metaverse.

This paper addresses the significant challenges in 3D Semantic Scene Graph (3DSSG) prediction, essential for understanding complex 3D environments. Traditional approaches, primarily using PointNet and Graph Convolutional Networks , struggle with effectively extracting multi-grained features from intricate 3D scenes , largely due to a focus on global scene processing and single-scale feature extraction. To overcome these limitations, we introduce Granular3D, a novel approach that shifts the focus towards multi-granularity analysis by predicting relation triplets from specific sub-scenes. One key is the Adaptive Instance Enveloping Method (AIEM), which establishes an approximate envelope structure around irregular instances, providing shape-adaptive local point cloud sampling, thereby comprehensively covering the contextual environments of instances. Moreover, Granular3D incorporates a Hierarchical Dual-Stage Network (HDSN), which differentiates and processes features of instances and their pairs at varying scales, leading to a targeted prediction of instance categories and their relationships. To advance the perception of sub-scene in HDSN, we design a Gather Point Transformer structure (GaPT) that enables the combinatorial interaction of local information from multiple point cloud sets, achieving a more comprehensive local contextual feature extraction. Extensive evaluations on the challenging 3DSSG benchmark demonstrate that our methods provide substantial improvements, establishing a new state-of-the-art in 3DSSG prediction, boosting the top-50 triplet accuracy by ＋2.8%.

Face swapping aims to transfer the identity of a source face to a target face image while preserving the target attributes (e.g., facial expression, head pose, illumination, and background). Most existing methods use a face recognition model to extract global features from the source face and directly fuse them with the target to generate a swapping result. However, identity-irrelevant attributes (e.g., hairstyle and facial appearances) contribute a lot to the recognition task, and thus swapping this task-specific feature inevitably interfuses source attributes with target ones. In this paper, we propose an identity-aware variational autoencoder (ID-VAE) based face swapping framework, dubbed VAFSwap, which learns disentangled identity and attribute representations for high-fidelity face swapping. In particular, we overcome the unpaired training barrier of VAE and impose a proxy identity on the latent space by exploiting the weak supervision from an auxiliary image set whose identity is averaged from multiple collected face images. To explicitly guide the identity fusion, we further devise an identity-associated matrix that corresponds different face regions with their identity representations to perform identity-related feature interactions. Finally, we incorporate spatial dimensions into the latent space and exploit the generative priors of a pre-trained face generator, allowing the effective elimination of noticeable swapping artifacts. Extensive experiments on the FaceForensics++ and CelebA-HQ datasets demonstrate that our method outperforms the state-of-the-art significantly.

In Visual Question Answering (VQA), addressing language prior bias, where models excessively rely on superficial correlations between questions and answers, is crucial. This issue becomes more pronounced in real-world applications with diverse domains and varied question-answer distributions during testing. To tackle this challenge, Test-time Adaptation (TTA) has emerged, allowing pre-trained VQA models to adapt using unlabeled test samples. Current state-of-the-art models select reliable test samples based on fixed entropy thresholds and employ self-supervised debiasing techniques. However, these methods struggle with diverse answer spaces linked to different question types and may fail to identify biased samples that still leverage relevant visual context. In this paper, we propose Question type-guided Entropy Minimization and Debiasing (QED) as a solution for test-time VQA model adaptation. Our approach involves adaptive entropy minimization based on question types to improve the identification of fine-grained and unreliable samples. Additionally, we generate negative samples for each test sample and label them as biased if their answer entropy change rate significantly differs from positive test samples, subsequently removing them. We evaluate our approach on two public benchmarks, VQA-CP v2, and VQA-CP v1, and achieve new state-of-the-art results, with overall accuracy rates of 48.13% and 46.18%, respectively.

Reversible face anonymization, unlike traditional face pixelization, seeks to replace sensitive identity information in facial images with synthesized alternatives, preserving privacy without sacrificing image clarity. Traditional methods, such as encoder-decoder networks, often result in significant loss of facial details due to their limited learning capacity. Additionally, relying on latent manipulation in pre-trained GANs can lead to changes in ID-irrelevant attributes, adversely affecting data utility due to GAN inversion inaccuracies. This paper introduces G2Face, which leverages both generative and geometric priors to enhance identity manipulation, achieving high-quality reversible face anonymization without compromising data utility. We utilize a 3D face model to extract geometric information from the input face, integrating it with a pre-trained GAN-based decoder. This synergy of generative and geometric priors allows the decoder to produce realistic anonymized faces with consistent geometry. Moreover, multi-scale facial features are extracted from the original face and combined with the decoder using our novel identity-aware feature fusion blocks (IFF). This integration enables precise blending of the generated facial patterns with the original ID-irrelevant features, resulting in accurate identity manipulation. Extensive experiments demonstrate that our method outperforms existing state-of-the-art techniques in face anonymization and recovery, while preserving high data utility. Code is available athttps://github.com/Harxis/G2Face.

The degradation of printed photographs due to inadequate preservation is a major problem that can be addressed through deep learning-based restoration methods. However, these methods are often limited by their reliance on annotated data, making them less effective for new domains with limited training samples. In this paper, we propose a semi-supervised old photo restoration network that employs a continuous important sample mining strategy. Specifically, we explore the learning potential of limited data from three aspects: correcting imbalanced data distribution, assigning significant pseudo labels, and learning from unlabeled data. First, we coordinate a random mask augmented strategy with the Double-consistency Alignment method to address the unbalanced damaged category (scratched damage is more prevalent than other artifact types). Second, we develop a novel Perceptual-aware Pseudo-label Propagation method that selects initial recovered results as reliable pseudo-labels to continuously expand the sample pool. Lastly, we propose a Damage-augmented Contrastive Learning method that constructs positive, anchor, and negative samples within a semi-supervised framework to mine correlations of unlabeled data more effectively. To evaluate our approach, we introduce the Old Photo Detection Dataset (OPDD) and the Old Photo Restoration Dataset (OPRD), both of which consist of 563 (6,179 augmented) photo pairs recovered by professional artists. Our extensive experiments show that our approach significantly outperforms existing methods. Furthermore, we demonstrate the effectiveness of our approach by training an external old photographic plate restoration network using the deuterogenic old photographic film dataset and obtaining promising results.

Learning discriminative and robust representations is important for facial expression recognition (FER) due to subtly different emotional faces and their subjective annotations. Previous works usually address one representation solely because these two goals seem to be contradictory for optimization. Their performances inevitably suffer from challenges from the other representation. In this article, by considering this problem from two novel perspectives, we demonstrate that discriminative and robust representations can be learned in a unified approach, i.e., DR-FER, and mutually benefit each other. Moreover, we make it with the supervision from only original annotations. Specifically, to learn discriminative representations, we propose performing masked image modeling (MIM) as an auxiliary task to force our network to discover expression-related facial areas. This is the first attempt to employ MIM to explore discriminative patterns in a self-supervised manner. To extract robust representations, we present a category-aware self-paced learning schedule to mine high-quality annotated (easy) expressions and incorrectly annotated (hard) counterparts. We further introduce a retrieval similarity-based relabeling strategy to correct hard expression annotations, exploiting them more effectively. By enhancing the discrimination ability of the FER classifier as a bridge, these two learning goals significantly strengthen each other. Extensive experiments on several popular benchmarks demonstrate the superior performance of our DR-FER. Moreover, thorough visualizations and extra experiments on manually annotation-corrupted datasets show that our approach successfully accomplishes learning both discriminative and robust representations simultaneously.

Photorealistic multiview face synthesis from a single image is a challenging problem. Existing works mainly learn a texture mapping model from the source to the target faces. However, they rarely consider the geometric constraints on the internal deformation arising from pose variations, which causes a high level of uncertainty in face pose modeling, and hence, produces inferior results for large pose variations. Moreover, current methods typically suffer from undesired facial details loss due to the adoption of the de-facto standard encoder-decoder architecture without any skip connections (SCs). In this article, we directly learn and exploit geometric constraints and propose a fully deformable network to simultaneously model the deformations of both landmarks and faces for face synthesis. Specifically, our model consists of two parts: a deformable landmark learning network (DLLN) and a gated deformable face synthesis network (GDFSN). The DLLN converts an initial reference landmark to an individual-specific target landmark as delicate pose guidance for face rotation. The GDFSN adopts a dual-stream structure, with one stream estimating the deformation of two views in the form of convolution offsets according to the source pose and the converted target pose, and the other leveraging the predicted deformation offsets to create the target face. In this way, individual-aware pose changes are explicitly modeled in the face generator to cope with geometric transformation, by adaptively focusing on pertinent regions of the source face. To compensate for offset estimation errors, we introduce a soft-gating mechanism for adaptive fusion between deformable features and primitive features. Additionally, a pose-aligned SC (PASC) is tailored to propagate low-level input features to the appropriate positions in the output features for further enhancing the facial details and identity preservation. Extensive experiments on six benchmarks show that our approach performs favorably against the state-of-the-arts, especially with large pose changes. Code is available at https://github.com/cschengxu/FDFace.

Nighttime semantic segmentation is an important but challenging research problem for autonomous driving. The major challenges lie in the small objects or regions from the under-/over-exposed areas or suffer from motion blur caused by the camera deployed on moving vehicles. To resolve this, we propose a novel hard-class-aware module that bridges the main network for full-class segmentation and the hard-class network for segmenting aforementioned hard-class objects. In specific, it exploits the shared focus of hard-class objects from the dual-stream network, enabling the contextual information flow to guide the model to concentrate on the pixels that are hard to classify. In the end, the estimated hard-class segmentation results will be utilized to infer the final results via an adaptive probabilistic fusion refinement scheme. Moreover, to overcome over-smoothing and noise caused by extreme exposures, our model is modulated by a carefully crafted pretext task of constructing an exposure-aware semantic gradient map, which guides the model to faithfully perceive the structural and semantic information of hard-class objects while mitigating the negative impact of noises and uneven exposures. In experiments, we demonstrate that our unique network design leads to superior segmentation performance over existing methods, featuring the strong ability of perceiving hard-class objects under adverse conditions.

Converting a human portrait to anime style is a desirable but challenging problem. Existing methods fail to resolve this problem due to the large inherent gap between two domains that cannot be overcome by a simple direct mapping. For this reason, these methods struggle to preserve the appearance features in the original photo. In this article, we discover an intermediate domain, the coser portrait (portraits of humans costuming as anime characters), that helps bridge this gap. It alleviates the learning ambiguity and loosens the mapping difficulty in a progressive manner. Specifically, we start from learning the mapping between coser and anime portraits, and present a proxy-guided domain adaptation learning scheme with three progressive adaptation stages to shift the initial model to the human portrait domain. In this way, our model can generate visually pleasant anime portraits with well-preserved appearances given the human portrait. Our model adopts a disentangled design by breaking down the translation problem into two specific subtasks of face deformation and portrait stylization. This further elevates the generation quality. Extensive experimental results show that our model can achieve visually compelling translation with better appearance preservation and perform favorably against the existing methods both qualitatively and quantitatively. Our code and datasets are available at https://github.com/NeverGiveU/PDA-Translation.

Layout generation plays a crucial role in graphic design intelligence. One important characteristic of the graphic layouts is that they usually follow certain design principles. For example, the principle of repetition emphasizes the reuse of similar visual elements throughout the design. To generate a layout, previous works mainly attempt at predicting the absolute value of bounding box for each element, where such target representation has hidden the information of higher-order design operations like repetition (e.g. copy the size of the previously generated element). In this paper, we introduce a novel action schema to encode these operations for better modeling the generation process. Instead of predicting the bounding box values, our approach autoregressively outputs the intermediate action sequence, which can then be deterministically converted to the final layout. We achieve state-of-the-art performances on three datasets. Both automatic and human evaluations show that our approach generates high-quality and diverse layouts. Furthermore, we revisit the commonly used evaluation metric FID adapted in this task, and observe that previous works use different settings to train the feature extractor for obtaining real/generated data distribution, which leads to inconsistent conclusions. We conduct an in-depth analysis on this metric and settle for a more robust and reliable evaluation setting. Code is available at this website.

Previous Knowledge Distillation based efficient image retrieval methods employ a lightweight network as the stu-dent model for fast inference. However, the lightweight stu-dent model lacks adequate representation capacity for effective knowledge imitation during the most critical early training period, causing final performance degeneration. To tackle this issue, we propose a Capacity Dynamic Distillation framework, which constructs a student model with editable representation capacity. Specifically, the employed student model is initially a heavy model to fruitfully learn distilled knowledge in the early training epochs, and the stu-dent model is gradually compressed during the training. To dynamically adjust the model capacity, our dynamic frame-work inserts a learnable convolutional layer within each residual block in the student model as the channel importance indicator. The indicator is optimized simultaneously by the image retrieval loss and the compression loss, and a retrieval- guided gradient resetting mechanism is proposed to release the gradient conflict. Extensive experiments show that our method has superior inference speed and accu-racy, e.g., on the VeRi-776 dataset, given the ResNet101 as a teacher, our method saves 67.13% model parameters and 65.67% FLOPs without sacrificing accuracy. Code is avail-able at https://github.com/SCY-X/Capacity_Dynamic_Distillation.

Image generation relies on massive training data that can hardly produce diverse images of an unseen category according to a few examples. In this paper, we address this dilemma by projecting sparse few-shot samples into a continuous latent space that can potentially generate infinite unseen samples. The rationale behind is that we aim to locate a centroid latent position in a conditional StyleGAN, where the corresponding output image on that centroid can maximize the similarity with the given samples. Although the given samples are unseen for the conditional StyleGAN, we assume the neighboring latent subspace around the centroid belongs to the novel category, and therefore introduce two latent subspace optimization objectives. In the first one we use few-shot samples as positive anchors of the novel class, and adjust the StyleGAN to produce the corresponding results with the new class label condition. The second objective is to govern the generation process from the other way around, by altering the centroid and its surrounding latent subspace for a more precise generation of the novel class. These reciprocal optimization objectives inject a novel class into the StyleGAN latent subspace, and therefore new unseen samples can be easily produced by sampling images from it. Extensive experiments demonstrate superior few-shot generation performances compared with state-of-the-art methods, especially in terms of diversity and generation quality. Code is available at https://github.com/chansey0529/LSO.

Considering the ill-posed nature, contrastive regularization has been developed for single image dehazing, introducing the information from negative images as a lower bound. However, the contrastive samples are non-consensual, as the negatives are usually represented distantly from the clear (i.e., positive) image, leaving the solution space still under-constricted. Moreover, the interpretability of deep dehazing models is underexplored towards the physics of the hazing process. In this paper, we propose a novel curricular contrastive regularization targeted at a consensual contrastive space as opposed to a non-consensual one. Our negatives, which provide better lower-bound constraints, can be assembled from 1) the hazy image, and 2) corresponding restorations by other existing methods. Further, due to the different similarities between the embeddings of the clear image and negatives, the learning difficulty of the multiple components is intrinsically imbalanced. To tackle this issue, we customize a curriculum learning strategy to reweight the importance of different negatives. In addition, to improve the interpretability in the feature space, we build a physics-aware dual-branch unit according to the atmospheric scattering model. With the unit, as well as curricular contrastive regularization, we establish our dehazing network, named C2PNet. Extensive experiments demonstrate that our C2PNet significantly outperforms state-of-the-art methods, with extreme PSNR boosts of 3.94dB and 1.50dB, respectively, on SOTS-indoor and SOTS-outdoor datasets. Code is available at https://github.com/YuZheng9/C2PNet.

This paper aims to resolve the challenging problem of wide-angle novel view synthesis from a single image, a.k.a. wide-angle 3D photography. Existing approaches rely on local context and treat them equally to inpaint occluded RGB and depth regions, which fail to deal with large-region occlusion (i.e., observing from an extreme angle) and foreground layers might blend into background inpainting. To address the above issues, we propose Diffuse3D which employs a pre-trained diffusion model for global synthesis, while amending the model to activate depth-aware inference. Our key insight is to alter the convolution mechanism in the denoising process. We inject depth information into the denoising convolution operation with bilateral kernels, i.e., a depth kernel and a spatial kernel, to consider layered correlations among pixels. In this way, foreground regions are overlooked in background inpainting and only pixels close in depth are leveraged. On the other hand, we propose a global-local balancing approach to maximize both contextual understandings. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods in novel view synthesis, especially in wide-angle scenarios. More importantly, our method does not require any training and is a plug-and-play module that can be integrated with any diffusion model. Our code can be found at https://github.com/yutaojiang1/Diffuse3D.

Removing moiré patterns from videos recorded on screens or complex textures is known as video demoiréing. It is a challenging task as both structures and textures of an image usually exhibit strong periodic patterns, which thus are easily confused with moiré patterns and can be significantly erased in the removal process. By interpreting video demoiréing as a multi-frame decomposition problem, we propose a compact invertible dyadic network called CIDNet that progressively decouples latent frames and the moiré patterns from an input video sequence. Using a dyadic cross-scale coupling structure with coupling layers tailored for multi-scale processing, CIDNet aims at disentangling the features of image patterns from that of moiré patterns at different scales, while retaining all latent image features to facilitate reconstruction. In addition, a compressed form for the network’s output is introduced to reduce computational complexity and alleviate overfitting. The experiments show that CIDNet outperforms existing methods and enjoys the advantages in model size and computational efficiency.

GAN inversion is indispensable for applying the powerful editability of GAN to real images. However, existing methods invert video frames individually often leading to undesired inconsistent results over time. In this paper, we propose a unified recurrent framework, named Recurrent vIdeo GAN Inversion and eDiting (RIGID), to explicitly and simultaneously enforce temporally coherent GAN inversion and facial editing of real videos. Our approach models the temporal relations between current and previous frames from three aspects. To enable a faithful real video reconstruction, we first maximize the inversion fidelity and consistency by learning a temporal compensated latent code. Second, we observe incoherent noises lie in the high-frequency domain that can be disentangled from the latent space. Third, to remove the inconsistency after attribute manipulation, we propose an in-between frame composition constraint such that the arbitrary frame must be a direct composite of its neighboring frames. Our unified framework learns the inherent coherence between input frames in an end-to-end manner, and therefore it is agnostic to a specific attribute and can be applied to arbitrary editing of the same video without re-training. Extensive experiments demonstrate that RIGID outperforms state-of-the-art methods qualitatively and quantitatively in both inversion and editing tasks. The deliverables can be found in https://cnnlstm.github.io/RIGID.

Video inpainting aims at filling in missing regions of a video. However, when dealing with dynamic scenes with camera or object movements, annotating the inpainting target becomes laborious and impractical. In this paper, we resolve the one-shot video inpainting problem in which only one annotated first frame is provided. A naive solution is to propagate the initial target to the other frames with techniques like object tracking. In this context, the main obstacles are the unreliable propagation and the partially inpainted artifacts due to the inaccurate mask. For the former problem, we propose curricular inactivation to replace the hard masking mechanism for indicating the in-painting target, which is robust to erroneous predictions in long-term video inpainting. For the latter, we explore the properties of inpainting residue and present an online residue removal method in an iterative detect-and-refine manner. Extensive experiments on several real-world datasets demonstrate the quantitative and qualitative superiorities of our proposed method in one-shot video inpainting. More importantly, our method is extremely flexible that can be integrated with arbitrary traditional inpainting models, activating them to perform the reliable one-shot video inpainting task. Video demonstrations can be found in our supplement, and our code can be found at https://github.com/Arise-zwy/CIRI.

Due to domain shift, a large performance drop is usually observed when a trained crowd counting model is deployed in the wild. While existing domain-adaptive crowd counting methods achieve promising results, they typically regard each crowd image as a whole and reduce domain discrepancies in a holistic manner, thus limiting further improvement of domain adaptation performance. To this end, we propose to untangle domain-invariant crowd and domain-specific background from crowd images and design a fine-grained domain adaption method for crowd counting. Specifically, to disentangle crowd from background, we propose to learn crowd segmentation from point-level crowd counting annotations in a weakly-supervised manner. Based on the derived segmentation, we design a crowd-aware domain adaptation mechanism consisting of two crowd-aware adaptation modules, i.e., Crowd Region Transfer (CRT) and Crowd Density Alignment (CDA). The CRT module is designed to guide crowd features transfer across domains beyond background distractions. The CDA module dedicates to regularising target-domain crowd density generation by its own crowd density distribution. Our method outperforms previous approaches consistently in the widely-used adaptation scenarios.

Multi-view (or -modality) representation learning aims to understand the relationships between different view representations. Existing methods disentangle multi-view representations into consistent and view-specific representations by introducing strong inductive biases, which can limit their generalization ability. In this paper, we propose a novel multi-view representation disentangling method that aims to go beyond inductive biases, ensuring both interpretability and generalizability of the resulting representations. Our method is based on the observation that discovering multi-view consistency in advance can determine the disentangling information boundary, leading to a decoupled learning objective. We also found that the consistency can be easily extracted by maximizing the transformation invariance and clustering consistency between views. These observations drive us to propose a two-stage framework. In the first stage, we obtain multi-view consistency by training a consistent encoder to produce semantically-consistent representations across views as well as their corresponding pseudo-labels. In the second stage, we disentangle specificity from comprehensive representations by minimizing the upper bound of mutual information between consistent and comprehensive representations. Finally, we reconstruct the original data by concatenating pseudo-labels and view-specific representations. Our experiments on four multi-view datasets demonstrate that our proposed method outperforms 12 comparison methods in terms of clustering and classification performance. The visualization results also show that the extracted consistency and specificity are compact and interpretable. Our code can be found at https://github.com/Guanzhou-Ke/DMRIB.

Non-photorealistic videos are in demand with the wave of the metaverse, but lack of sufficient research studies. This work aims to take a step forward to understand how humans perceive non-photorealistic videos with eye fixation (i.e., saliency detection), which is critical for enhancing media production, artistic design, and game user experience. To fill in the gap of missing a suitable dataset for this research line, we present NPF-200, the first large-scale multi-modal dataset of purely non-photorealistic videos with eye fixations. Our dataset has three characteristics: 1) it contains soundtracks that are essential according to vision and psychological studies; 2) it includes diverse semantic content and videos are of high-quality; 3) it has rich motions across and within videos. We conduct a series of analyses to gain deeper insights into this task and compare several state-of-the-art methods to explore the gap between natural images and non-photorealistic data. Additionally, as the human attention system tends to extract visual and audio features with different frequencies, we propose a universal frequency-aware multi-modal non-photorealistic saliency detection model called NPSNet, demonstrating the state-of-the-art performance of our task. The results uncover strengths and weaknesses of multi-modal network design and multi-domain training, opening up promising directions for future works. Our dataset and code can be found at https://github.com/Yangziyu/NPF200

Domain adaptation is commonly employed in crowd counting to bridge the domain gaps between different datasets. However, existing domain adaptation methods tend to focus on inter-dataset differences while overlooking the intra-differences within the same dataset, leading to additional learning ambiguities. These domain-agnostic factors,e.g., density, surveillance perspective, and scale, can cause significant in-domain variations, and the misalignment of these factors across domains can lead to a drop in performance in cross-domain crowd counting. To address this issue, we propose a Domain-agnostically Aligned Optimal Transport (DAOT) strategy that aligns domain-agnostic factors between domains. The DAOT consists of three steps. First, individual-level differences in domain-agnostic factors are measured using structural similarity (SSIM). Second, the optimal transfer (OT) strategy is employed to smooth out these differences and find the optimal domain-to-domain misalignment, with outlier individuals removed via a virtual "dustbin'' column. Third, knowledge is transferred based on the aligned domain-agnostic factors, and the model is retrained for domain adaptation to bridge the gap across domains. We conduct extensive experiments on five standard crowd-counting benchmarks and demonstrate that the proposed method has strong generalizability across diverse datasets. Our code will be available at: https://github.com/HopooLinZ/DAOT/.

Abstract Adding screentone to initial line drawings is a crucial step for manga generation, but is a tedious and human‐laborious task. In this work, we propose a novel data‐driven method aiming to transfer the screentone pattern from a reference manga image. This not only ensures the quality, but also adds controllability to the generated manga results. The reference‐based screentone translation task imposes several unique challenges. Since manga image often contains multiple screentone patterns interweaved with line drawing, as an abstract art, this makes it even more difficult to extract disentangled style code from the reference. Also, finding correspondence for mapping between the reference and the input line drawing without any screentone is hard. As screentone contains many subtle details, how to guarantee the style consistency to the reference remains challenging. To suit our purpose and resolve the above difficulties, we propose a novel Reference‐based Screentone Transfer Network (RSTN). We encode the screentone style through a 1D stylegram. A patch correspondence loss is designed to build a similarity mapping function for guiding the translation. To mitigate the generated artefacts, a pattern regularization loss is introduced in the patch‐level. Through extensive experiments and a user study, we have demonstrated the effectiveness of our proposed model.

Traditional monocular depth estimation assumes that all objects are reliably visible in the RGB color domain. However, this is not always the case as more and more buildings are decorated with transparent glass walls. This problem has not been explored due to the difficulties in annotating the depth levels of glass walls, as commercial depth sensors cannot provide correct feedbacks on transparent objects. Furthermore, estimating depths from transparent glass walls requires the aids of surrounding context, which has not been considered in prior works. To cope with this problem, we introduce the first Glass Walls Depth Dataset (GW-Depth dataset). We annotate the depth levels of transparent glass walls by propagating the context depth values within neighboring flat areas, and the glass segmentation mask and instance level line segments of glass edges are also provided. On the other hand, a tailored monocular depth estimation method is proposed to fully activate the glass wall contextual understanding. First, we propose to exploit the glass structure context by incorporating the structural prior knowledge embedded in glass boundary line segment detections. Furthermore, to make our method adaptive to scenes without structure context where the glass boundary is either absent in the image or too narrow to be recognized, we propose to derive a reflection context by utilizing the depth reliable points sampled according to the variance between two depth estimations from different resolutions. High-resolution depth is thus estimated by the weighted summation of depths by those reliable points. Extensive experiments are conducted to evaluate the effectiveness of the proposed dual context design. Superior performances of our method is also demonstrated by comparing with state-of-the-art methods. We present the first feasible solution for monocular depth estimation in the presence of glass walls, which can be widely adopted in autonomous navigation.

Labeling is onerous for crowd counting as it should annotate each individual in crowd images. Recently, several methods have been proposed for semi-supervised crowd counting to reduce the labeling efforts. Given a limited labeling budget, they typically select a few crowd images and densely label all individuals in each of them. Despite the promising results, we argue the None-or-All labeling strategy is suboptimal as the densely labeled individuals in each crowd image usually appear similar while the massive unlabeled crowd images may contain entirely diverse individuals. To this end, we propose to break the labeling chain of previous methods and make the first attempt to reduce spatial labeling redundancy for semi-supervised crowd counting. First, instead of annotating all the regions in each crowd image, we propose to annotate the representative ones only. We analyze the region representativeness from both vertical and horizontal directions of initially estimated density maps, and formulate them as cluster centers of Gaussian Mixture Models. Additionally, to leverage the rich unlabeled regions, we exploit the similarities among individuals in each crowd image to directly supervise the unlabeled regions via feature propagation instead of the error-prone label propagation employed in the previous methods. In this way, we can transfer the original spatial labeling redundancy caused by individual similarities to effective supervision signals on the unlabeled regions. Extensive experiments on the widely-used benchmarks demonstrate that our method can outperform previous best approaches by a large margin.

Printed photographs can be easily warped, wrinkled, and even deteriorated over time. Existing methods treat the restoration of scratches as a pure inpainting problem that neglects the underlying corrupted contextual knowledge. However, important underlying contents are hidden behind the scratches, which are essential hints for producing a semantically consistent result. Motivated by this insight, we explore how to harmonize the scratch-free features and noisy but essential scratch features to produce a visually consistent restoration. Specifically, in this paper, we propose an automatic retouching approach for scratched photographs with the aid of scratch/background context. We explicitly process scratch and background context in two stages. In the first stage, we mainly extract global scratch features, while the mask is introduced in the second stage to filter out and inpaint the scratches. Both contexts are carefully reciprocated for a faithful restoration. Particularly, we propose a Scratch Contextual Assisted Module (SCAM) to adaptively learn texture within the detected mask. This module utilizes the distance between the scratch mask-out feature and scratch encoder feature for modeling the pixel-wise correspondence, which determines the importance of the encoder feature within the scratch mask. Furthermore, to facilitate the evaluation of scratch restoration methods, we create two new scratched photo datasets which have 238 scratch/scratch-free photo pairs to promote the development in the scratch restoration field, namely Old Scratched Photo Dataset (OSPD) and Modern Scratched Photo Dataset (MSPD). Extensive experimental results on the proposed datasets demonstrate that our model outperforms existing methods. To extend the application, we also perform the proposed method on video samples and obtain visual-pleasing results. The code can be found athttps://github.com/cwyyt/Contextual-assisted-Scratched-Photo-Restoration.

Anime is an abstract art form that is substantially different from the human portrait, leading to a challenging misaligned image translation problem that is beyond the capability of existing methods. This can be boiled down to a highly ambiguous unconstrained translation between two domains. To this end, we design a new anime translation framework by deriving the prior knowledge of a pre-trained StyleGAN model. We introduce disentangled encoders to separately embed structure and appearance information into the same latent code, governed by four tailored losses. Moreover, we develop a FaceBank aggregation method that leverages the generated data of the StyleGAN, anchoring the prediction to produce in-domain animes. To empower our model and promote the research of anime translation, we propose the first anime portrait parsing dataset, Danbooru-Parsing , containing 4,921 densely labeled images across 17 classes. This dataset connects the face semantics with appearances, enabling our new constrained translation setting. We further show the editability of our results, and extend our method to manga images, by generating the first manga parsing pseudo data. Extensive experiments demonstrate the values of our new dataset and method, resulting in the first feasible solution on anime translation.

Person Image Synthesis aims at transferring the appearance of the source person image into a target pose. Existing methods cannot handle large pose variations and therefore suffer from two critical problems: (1) synthesis distortion due to the entanglement of pose and appearance information among different body components and (2) failure in preserving original semantics (e.g., the same outfit). In this article, we explicitly address these two problems by proposing a Pose- and Attribute-consistent Person Image Synthesis Network (PAC-GAN). To reduce pose and appearance matching ambiguity, we propose a component-wise transferring model consisting of two stages. The former stage focuses only on synthesizing target poses, while the latter renders target appearances by explicitly transferring the appearance information from the source image to the target image in a component-wise manner. In this way, source-target matching ambiguity is eliminated due to the component-wise disentanglement of pose and appearance synthesis. Second, to maintain attribute consistency, we represent the input image as an attribute vector and impose a high-level semantic constraint using this vector to regularize the target synthesis. Extensive experimental results on the DeepFashion dataset demonstrate the superiority of our method over the state of the art, especially for maintaining pose and attribute consistencies under large pose variations.

The inductive bias of vision transformers is more relaxed that cannot work well with insufficient data. Knowledge distillation is thus introduced to assist the training of transformers. Unlike previous works, where merely heavy convolution-based teachers are provided, in this paper, we delve into the influence of models inductive biases in knowledge distillation (e.g., convolution and involution). Our key observation is that the teacher accuracy is not the dominant reason for the student accuracy, but the teacher inductive bias is more important. We demonstrate that lightweight teachers with different architectural inductive biases can be used to co-advise the student transformer with outstanding performances. The rationale behind is that models designed with different inductive biases tend to focus on diverse patterns, and teachers with different inductive biases attain various knowledge despite being trained on the same dataset. The diverse knowledge provides a more precise and comprehensive description of the data and compounds and boosts the performance of the student during distillation. Furthermore, we propose a token inductive bias alignment to align the inductive bias of the token with its target teacher model. With only lightweight teachers provided and using this cross inductive bias distillation method, our vision transformers (termed as CiT) outperform all previous vision transformers (ViT) of the same architecture on ImageNet. Moreover, our small size model CiT-SAK further achieves 82.7% Top-1 accuracy on ImageNet without modifying the attention module of the ViT. Code is available at https://github.com/OliverRensu/co-advise.

Data mixing (e.g., Mixup, Cutmix, ResizeMix) is an essential component for advancing recognition models. In this paper, we focus on studying its effectiveness in the self-supervised setting. By noticing the mixed images that share the same source images are intrinsically related to each other, we hereby propose SDMP, short for Simple Data Mixing Prior, to capture this straightforward yet essential prior, and position such mixed images as additional positive pairs to facilitate self-supervised representation learning. Our experiments verify that the proposed SDMP enables data mixing to help a set of self-supervised learning frameworks (e.g., MoCo) achieve better accuracy and out-of-distribution robustness. More notably, our SDMP is the first method that successfully leverages data mixing to improve (rather than hurt) the performance of Vision Transformers in the self-supervised setting. Code is publicly available at https://github.com/OliverRensu/SDMP.

Recent Vision Transformer (ViT) models have demonstrated encouraging results across various computer vision tasks, thanks to its competence in modeling long-range de-pendencies of image patches or tokens via self-attention. These models, however, usually designate the similar receptive fields of each token feature within each layer. Such a constraint inevitably limits the ability of each self-attention layer in capturing multi-scale features, thereby leading to performance degradation in handling images with multiple objects of different scales. To address this issue, we propose a novel and generic strategy, termed shunted self-attention (SSA), that allows ViTs to model the attentions at hybrid scales per attention layer. The key idea of SSA is to inject heterogeneous receptive field sizes into tokens: before computing the self-attention matrix, it selectively merges tokens to represent larger object features while keeping certain tokens to preserve fine-grained features. This novel merging scheme enables the self-attention to learn relationships between objects with different sizes, and simultaneously reduces the token numbers and the computational cost. Extensive experiments across various tasks demonstrate the superiority of SSA. Specifically, the SSA-based transformer achieve 84.0% Top-1 accuracy and out-performs the state-of-the-art Focal Transformer on Ima-geNet with only half of the model size and computation cost, and surpasses Focal Transformer by 1.3 mAP on COCO and 2.9 mIOU on ADE20K under similar parameter and computation cost. Code has been released at https://github.com/OliverRensulShunted-Transformer.

We present a novel high-resolution face swapping method using the inherent prior knowledge of a pre-trained GAN model. Although previous research can leverage generative priors to produce high-resolution results, their quality can suffer from the entangled semantics of the latent space. We explicitly disentangle the latent semantics by utilizing the progressive nature of the generator, deriving structure at-tributes from the shallow layers and appearance attributes from the deeper ones. Identity and pose information within the structure attributes are further separated by introducing a landmark-driven structure transfer latent direction. The disentangled latent code produces rich generative features that incorporate feature blending to produce a plausible swapping result. We further extend our method to video face swapping by enforcing two spatio-temporal constraints on the latent space and the image space. Extensive experiments demonstrate that the proposed method outperforms state-of-the-art image/video face swapping methods in terms of hallucination quality and consistency. Code can be found at: https://github.com/cnnlstm/FSLSD_HiRes.

This paper presents a Generative prior ReciprocAted Invertible rescaling Network (GRAIN) for generating faithful high-resolution (HR) images from low-resolution (LR) invertible images with an extreme upscaling factor (64×). Previous researches have leveraged the prior knowledge of a pretrained GAN model to generate high-quality upscaling results. However, they fail to produce pixel-accurate results due to the highly ambiguous extreme mapping process. We remedy this problem by introducing a reciprocated invertible image rescaling process, in which high-resolution information can be delicately embedded into an invertible low-resolution image and generative prior for a faithful HR reconstruction. In particular, the invertible LR features not only carry significant HR semantics, but also are trained to predict scale-specific latent codes, yielding a preferable utilization of generative features. On the other hand, the enhanced generative prior is re-injected to the rescaling process, compensating the lost details of the invertible rescaling. Our reciprocal mechanism perfectly integrates the advantages of invertible encoding and generative prior, leading to the first feasible extreme rescaling solution. Extensive experiments demonstrate superior performance against state-of-the-art upscaling methods. Code is available at https://github.com/cszzx/GRAIN.

We propose a simple yet effective technique that significantly improves the performance of the current state-of-the-art background matting model without compromising its original speed. We achieve this by carefully exciting the proper neural activations using an excitation map in the training phase and performing recursive inference in the testing phase. To avoid being over-reliant on perfect excitations, we follow the idea of curriculum learning to divide the training phase into three easy-to-hard stages and gradually shift the excitation map from GT alpha matte to pseudo GT alpha matte. In the testing phase, we propose a recursive inference mechanism that uses the output alpha matte as the excitation map to further refine the output alpha matte. Our method is a simple plug-in for arbitrary matting models. Compared with the original ones, the enhanced models alleviate the problem of performance degradation with complex background and thus boosts the matting accuracy.

In this paper, we introduce a novel yet challenging research problem, interactive crowd video generation, committed to producing diverse and continuous crowd video, and relieving the difficulty of insufficient annotated real-world datasets in crowd analysis. Our goal is to recursively generate realistic future crowd video frames given few context frames, under the user-specified guidance, namely individual positions of the crowd. To this end, we propose a deep network architecture specifically designed for crowd video generation that is composed of two complementary modules, each of which combats the problems of crowd dynamic synthesis and appearance preservation respectively. Particularly, a spatio-temporal transfer module is proposed to infer the crowd position and structure from guidance and temporal information, and a point-aware flow prediction module is presented to preserve appearance consistency by flow-based warping. Then, the outputs of the two modules are integrated by a self-selective fusion unit to produce an identity-preserved and continuous video. Unlike previous works, we generate continuous crowd behaviors beyond identity annotations or matching. Extensive experiments show that our method is effective for crowd video generation. More importantly, we demonstrate the generated video can produce diverse crowd behaviors and be used for augmenting different crowd analysis tasks, i.e., crowd counting, anomaly detection, crowd video prediction. Code is available at https://github.com/Icep2020/CrowdGAN.

This paper proposes a novel pretext task to address the self-supervised video representation learning problem. Specifically, given an unlabeled video clip, we compute a series of spatio-temporal statistical summaries, such as the spatial location and dominant direction of the largest motion, the spatial location and dominant color of the largest color diversity along the temporal axis, etc. Then a neural network is built and trained to yield the statistical summaries given the video frames as inputs. In order to alleviate the learning difficulty, we employ several spatial partitioning patterns to encode rough spatial locations instead of exact spatial Cartesian coordinates. Our approach is inspired by the observation that human visual system is sensitive to rapidly changing contents in the visual field, and only needs impressions about rough spatial locations to understand the visual contents. To validate the effectiveness of the proposed approach, we conduct extensive experiments with four 3D backbone networks, i.e., C3D, 3D-ResNet, R(2+1)D and S3D-G. The results show that our approach outperforms the existing approaches across these backbone networks on four downstream video analysis tasks including action recognition, video retrieval, dynamic scene recognition, and action similarity labeling. The source code is publicly available at: https://github.com/laura-wang/video_repres_sts.

Multiview clustering seeks to partition objects via leveraging cross-view relations to provide a comprehensive description of the same objects. Most existing methods assume that different views are linear transformable or merely sampling from a common latent space. Such rigid assumptions betray reality, thus leading to unsatisfactory performance. To tackle the issue, we propose to learn both common and specific sampling spaces for each view to fully exploit their collaborative representations. The common space corresponds to the universal self-representation basis for all views, while the specific spaces are the view-specific basis accordingly. An iterative self-supervision scheme is conducted to strengthen the learned affinity matrix. The clustering is modeled by a convex optimization. We first solve its linear formulation by the popular scheme. Then, we employ the deep autoencoder structure to exploit its deep nonlinear formulation. The extensive experimental results on six real-world datasets demonstrate that the proposed model achieves uniform superiority over the benchmark methods.

We resolve the ill-posed alpha matting problem from a completely different perspective. Given an input portrait image, instead of estimating the corresponding alpha matte, we focus on the other end, to subtly enhance this input so that the alpha matte can be easily estimated by any existing matting models. This is accomplished by exploring the latent space of GAN models. It is demonstrated that interpretable directions can be found in the latent space and they correspond to semantic image transformations. We further explore this property in alpha matting. Particularly, we invert an input portrait into the latent code of StyleGAN, and our aim is to discover whether there is an enhanced version in the latent space which is more compatible with a reference matting model. We optimize multi-scale latent vectors in the latent spaces under four tailored losses, ensuring matting-specificity and subtle modifications on the portrait. We demonstrate that the proposed method can refine real portrait images for arbitrary matting models, boosting the performance of automatic alpha matting by a large margin. In addition, we leverage the generative property of StyleGAN, and propose to generate enhanced portrait data which can be treated as the pseudo GT. It addresses the problem of expensive alpha matte annotation, further augmenting the matting performance of existing models.

The state-of-the-art photo upsampling method, PULSE, demonstrates that a sharp, high-resolution (HR) version of a given low-resolution (LR) input can be obtained by exploring the latent space of generative models. However, mapping an extreme LR input (162) directly to an HR image (10242) is too ambiguous to preserve faithful local facial semantics. In this paper, we propose an enhanced upsampling approach, Pro-PULSE, that addresses the issues of semantic inconsistency and optimization complexity. Our idea is to learn an encoder that progressively constructs the HR latent codes in the extended$\mathcal {W}+$latent space of StyleGAN. This design divides the complex$64\times $upsampling problem into several steps, and therefore small-scale facial semantics can be inherited from one end to the other. In particular, we train two encoders, the base encoder maps latent vectors in$\mathcal {W}$space and serves as a foundation of the HR latent vector, while the second scale-specific encoder performed in$\mathcal {W}+$space gradually replaces the previous vector produced by the base encoder at each scale. This process produces intermediate side-outputs, which injects deep supervision into the training of encoder. Extensive experiments demonstrate superiorities over the latest latent space exploration methods, in terms of efficiency, quantitative quality metrics, and qualitative visual results.

Pixel art is a unique art style with the appearance of low resolution images. In this paper, we propose a data-driven pixelization method that can produce sharp and crisp cell effects with controllable cell sizes. Our approach overcomes the limitation of existing learning-based methods in cell size control by introducing a reference pixel art to explicitly regularize the cell structure. In particular, the cell structure features of the reference pixel art are used as an auxiliary input for the pixelization process, and for measuring the style similarity between the generated result and the reference pixel art. Furthermore, we disentangle the pixelization process into specific cell-aware and aliasing-aware stages, mitigating the ambiguities in joint learning of cell size, aliasing effect, and color assignment. To train our model, we construct a dedicated pixel art dataset and augment it with different cell sizes and different degrees of anti-aliasing effects. Extensive experiments demonstrate its superior performance over state-of-the-arts in terms of cell sharpness and perceptual expressiveness. We also show promising results of video game pixelization for the first time. Code and dataset are available at https://github.com/WuZongWei6/Pixelization.

Due to the challenges of densely compacted body parts, nonrigid clothing items, and severe overlap in crowd scenes, human parsing needs to focus more on multilevel feature representations compared to general scene parsing tasks. Based on this observation, we propose to introduce the auxiliary task of human mask and edge detection to facilitate human parsing. Different from human parsing, which exploits the discriminative features of each category, human mask and edge detection emphasizes the boundaries of semantic parsing regions and the difference between foreground humans and background clutter, which benefits the parsing predictions of crowd scenes and small human parts. Specifically, we extract human mask and edge labels from the human parsing annotations and train a shared encoder with three independent decoders for the three mutually beneficial tasks. Furthermore, the decoder feature maps of the human mask prediction branch are further exploited as attention maps, indicating human regions to facilitate the decoding process of human parsing and human edge detection. In addition to these auxiliary tasks, we further alleviate the problem of deformed clothing items under various human poses by tracking the deformation patterns with the deformable convolution. Extensive experiments show that the proposed method can achieve superior performance against state-of-the-art methods on both single and multiple human parsing datasets. Codes and trained models are available https://github.com/ViktorLiang/MGDAN .

Present studies have discovered that state-of-the-art deep learning models can be attacked by small but well-designed perturbations. Existing attack algorithms for the image captioning task is time-consuming, and their generated adversarial examples cannot transfer well to other models. To generate adversarial examples faster and stronger, we propose to learn the perturbations by a generative model that is governed by three novel loss functions. Image feature distortion loss is designed to maximize the encoded image feature distance between original images and the corresponding adversarial examples at the image domain, and local-global mismatching loss is introduced to separate the mapping encoding representation of the adversarial images and the ground true captions from a local and global perspective in the common semantic space as far as possible cross image and caption domain. Language diversity loss is to make the image captions generated by the adversarial examples as different as possible from the correct image caption at the language domain. Extensive experiments show that our proposed generative model can efficiently generate adversarial examples that successfully generalize to attack image captioning models trained on unseen large-scale datasets or with different architectures, or even the image captioning commercial service.

This paper tackles the task of Few-Shot Video Object Segmentation (FSVOS), i.e., segmenting objects in the query videos with certain class specified in a few labeled support images. The key is to model the relationship between the query videos and the support images for propagating the object information. This is a many-to-many problem and often relies on full-rank attention, which is computationally intensive. In this paper, we propose a novel Domain Agent Network (DAN), breaking down the full-rank attention into two smaller ones. We consider one single frame of the query video as the domain agent, bridging between the support images and the query video. Our DAN allows a linear space and time complexity as opposed to the original quadratic form with no loss of performance. In addition, we introduce a learning strategy by combining meta-learning with online learning to further improve the segmentation accuracy. We build a FSVOS benchmark on the Youtube-VIS dataset and conduct experiments to demonstrate that our method outperforms baselines on both computational cost and accuracy, achieving the state-of-the-art performance. Code is available at https://github.com/scutpaul/DANet.

Transferring makeup from the misaligned reference image is challenging. Previous methods overcome this barrier by computing pixel-wise correspondences between two images, which is inaccurate and computational-expensive. In this paper, we take a different perspective to break down the makeup transfer problem into a two-step extraction-assignment process. To this end, we propose a Style-based Controllable GAN model that consists of three components, each of which corresponds to target style-code encoding, face identity features extraction, and makeup fusion, respectively. In particular, a Part-specific Style Encoder encodes the component-wise makeup style of the reference image into a style-code in an intermediate latent space W. The style-code discards spatial information and therefore is invariant to spatial misalignment. On the other hand, the style-code embeds component-wise information, enabling flexible partial makeup editing from multiple references. This style-code, together with source identity features, is integrated into a Makeup Fusion Decoder equipped with multiple AdaIN layers to generate the final result. Our proposed method demonstrates great flexibility on makeup transfer by supporting makeup removal, shade-controllable makeup transfer, and part-specific makeup transfer, even with large spatial misalignment. Extensive experiments demonstrate the superiority of our approach over state-of-the-art methods. Code is available at https://github.com/makeuptransfer/SCGAN.

Lip reading aims to predict the spoken sentences from silent lip videos. Due to the fact that such a vision task usually performs worse than its counterpart speech recognition, one potential scheme is to distill knowledge from a teacher pretrained by audio signals. However, the latent domain gap between the cross-modal data could lead to a learning ambiguity and thus limits the performance of lip reading. In this paper, we propose a novel collaborative framework for lip reading, and two aspects of issues are considered: 1) the teacher should understand bi-modal knowledge to possibly bridge the inherent cross-modal gap; 2) the teacher should adjust teaching contents adaptively with the evolution of the student. To these ends, we introduce a trainable "master" network which ingests both audio signals and silent lip videos instead of a pretrained teacher. The master produces logits from three modalities of features: audio modality, video modality, and their combination. To further provide an interactive strategy to fuse these knowledge organically, we regularize the master with the task-specific feedback from the student, in which the requirement of the student is implicitly embedded. Meanwhile, we involve a couple of "tutor" networks into our system as guidance for emphasizing the fruitful knowledge flexibly. In addition, we incorporate a curriculum learning design to ensure a better convergence. Extensive experiments demonstrate that the proposed network outperforms the state-of-the-art methods on several benchmarks, including in both word-level and sentence-level scenarios.

Unsupervised video object segmentation (UVOS) aims at segmenting the primary objects in videos without any human intervention. Due to the lack of prior knowledge about the primary objects, identifying them from videos is the major challenge of UVOS. Previous methods often regard the moving objects as primary ones and rely on optical flow to capture the motion cues in videos, but the flow information alone is insufficient to distinguish the primary objects from the background objects that move together. This is because, when the noisy motion features are combined with the appearance features, the localization of the primary objects is misguided. To address this problem, we propose a novel reciprocal transformation network to discover primary objects by correlating three key factors: the intra-frame contrast, the motion cues, and temporal coherence of recurring objects. Each corresponds to a representative type of primary object, and our reciprocal mechanism enables an organic coordination of them to effectively remove ambiguous distractions from videos. Additionally, to exclude the information of the moving background objects from motion features, our transformation module enables to reciprocally transform the appearance features to enhance the motion features, so as to focus on the moving objects with salient appearance while removing the co-moving outliers. Experiments on the public benchmarks demonstrate that our model significantly outperforms the state-of-the-art methods. Code is available at https://github.com/OliverRensu/RTNet.

Generative adversarial networks (GANs) learn to map noise latent vectors to high-fidelity image outputs. It is found that the input latent space shows semantic correlations with the output image space. Recent works aim to interpret the latent space and discover meaningful directions that correspond to human interpretable image transformations. However, these methods either rely on explicit scores of attributes (e.g., memorability) or are restricted to binary ones (e.g., gender), which largely limits the applicability of editing tasks, especially for free-form artistic tasks like style/anime editing. In this paper, we propose an adversarial method, AdvStyle, for discovering interpretable directions in the absence of well-labeled scores or binary attributes. In particular, the proposed adversarial method simultaneously optimizes the discovered directions and the attribute assessor using the target attribute data as positive samples, while the generated ones being negative. In this way, arbitrary attributes can be edited by collecting positive data only, and the proposed method learns a controllable representation enabling manipulation of non-binary attributes like anime styles and facial characteristics. Moreover, the proposed learning strategy attenuates the entanglement between attributes, such that multi-attribute manipulation can be easily achieved without any additional constraint. Furthermore, we reveal several interesting semantics with the involuntarily learned negative directions. Extensive experiments on 9 anime attributes and 7 human attributes demonstrate the effectiveness of our adversarial approach qualitatively and quantitatively. Code is available at https://github.com/BERYLSHEEP/AdvStyle.

HD map reconstruction is crucial for autonomous driving. LiDAR-based methods are limited due to the deployed expensive sensors and time-consuming computation. Camera-based methods usually need to separately perform road segmentation and view transformation, which often causes distortion and the absence of content. To push the limits of the technology, we present a novel framework that enables reconstructing a local map formed by road layout and vehicle occupancy in the bird’s-eye view given a front-view monocular image only. In particular, we propose a cross-view transformation module, which takes the constraint of cycle consistency between views into account and makes full use of their correlation to strengthen the view transformation and scene understanding. Considering the relationship between vehicles and roads, we also design a context-aware discriminator to further refine the results. Experiments on public benchmarks show that our method achieves the state-of-the-art performance in the tasks of road layout estimation and vehicle occupancy estimation. Especially for the latter task, our model outperforms all competitors by a large margin. Furthermore, our model runs at 35 FPS on a single GPU, which is efficient and applicable for real-time panorama HD map reconstruction.

Ultra-high resolution image segmentation has raised increasing interests in recent years due to its realistic applications. In this paper, we innovate the widely used high-resolution image segmentation pipeline, in which an ultrahigh resolution image is partitioned into regular patches for local segmentation and then the local results are merged into a high-resolution semantic mask. In particular, we introduce a novel locality-aware contextual correlation based segmentation model to process local patches, where the relevance between local patch and its various contexts are jointly and complementarily utilized to handle the semantic regions with large variations. Additionally, we present a contextual semantics refinement network that associates the local segmentation result with its contextual semantics, and thus is endowed with the ability of reducing boundary artifacts and refining mask contours during the generation of final high-resolution mask. Furthermore, in comprehensive experiments, we demonstrate that our model outperforms other state-of-the-art methods in public benchmarks. Our released codes are available at https://github.com/liqiokkk/FCtL.

Existing GAN inversion methods are stuck in a paradox that the inverted codes can either achieve high-fidelity reconstruction, or retain the editing capability. Having only one of them clearly cannot realize real image editing. In this paper, we resolve this paradox by introducing consecutive images (e.g., video frames or the same person with different poses) into the inversion process. The rationale behind our solution is that the continuity of consecutive images leads to inherent editable directions. This inborn property is used for two unique purposes: 1) regularizing the joint inversion process, such that each of the inverted codes is semantically accessible from one of the other and fastened in an editable domain; 2) enforcing inter-image coherence, such that the fidelity of each inverted code can be maximized with the complement of other images. Extensive experiments demonstrate that our alternative significantly outperforms state-of-the-art methods in terms of reconstruction fidelity and editability on both the real image dataset and synthesis dataset. Furthermore, our method provides the first support of video-based GAN inversion and an interesting application of unsupervised semantic transfer from consecutive images. Source code can be found at: https://github.com/cnnlstm/InvertingGANs_with_ConsecutiveImgs.

Directly deploying a trained multi-modal classifier to a new environment usually leads to poor performance due to the well-known domain shift problem. Existing multi-modal domain adaptation methods treated each modality equally and optimize the sub-models of different modalities synchronously. However, as observed in this paper, the degrees of domain shift in different modalities are usually diverse. We propose a novel Differentiated Learning framework to make use of the diversity between multiple modalities for more effective domain adaptation. Specifically, we model the classifiers of different modalities as a group of teacher/student sub-models, and a novel Prototype based Reliability Measurement is presented to estimate the reliability of the recognition results made by each sub-model on the target domain. More reliable results are then picked up as teaching materials for all sub-models in the group. Considering the diversity of different modalities, each sub-model performs the Asynchronous Curriculum Learning by choosing the teaching materials from easy to hard measured by itself. Furthermore, a reliability-aware fusion scheme is proposed to combine all optimized sub-models to support final decision. Comprehensive experiments based on three multi-modal datasets with different learning tasks have been conducted, which show the superior performance of our model while comparing with state-of-the-art multi-modal domain adaptation models.

Unlike images, finding the desired video content in a large pool of videos is not easy due to the time cost of loading and watching. Most video streaming and sharing services provide the video preview function for a better browsing experience. In this paper, we aim to generate a video preview from a single image. To this end, we propose two cascaded networks, the motion embedding network and the motion expansion network. The motion embedding network aims to embed the spatio-temporal information into an embedded image, called video snapshot. On the other end, the motion expansion network is proposed to invert the video back from the input video snapshot. To hold the invertibility of motion embedding and expansion during training, we design four tailor-made losses and a motion attention module to make the network focus on the temporal information. In order to enhance the viewing experience, our expansion network involves an interpolation module to produce a longer video preview with a smooth transition. Extensive experiments demonstrate that our method can successfully embed the spatio-temporal information of a video into one "live" image, which can be converted back to a video preview. Quantitative and qualitative evaluations are conducted on a large number of videos to prove the effectiveness of our proposed method. In particular, statistics of PSNR and SSIM on a large number of videos show the proposed method is general, and it can generate a high-quality video from a single image.

Shadow removal is an important yet challenging task in image processing and computer vision. Existing methods are limited in extracting good global features due to the interference of shadow. And also, most of them ignore a fact that features inside and outside the shaded area should be treated disparately because of different semantics or materials. In this letter, we propose a novel deep neural network Mask-ShadowNet for shadow removal. The core of our approach is a well-designed masked adaptive instance normalization (MAdaIN) mechanism with embedded aligners that serves two goals: 1) producing hidden features that considering an illumination consistency of different regions. 2) treating the feature statistics of shadow and non-shadow areas discriminately based on the shadow mask. Experimental results demonstrate that the proposed model outperforms the state-of-the-art on the ISTD benchmark. Our code is available inhttps://github.com/penguinbing/Mask-ShadowNet.

Tumor metastases detection is of great importance for the treatment of breast cancer patients. Various CNN (convolutional neural network) based methods get excellent performance in object detection/segmentation. However, the detection of metastases in hematoxylin and eosin (H&E) stained whole-slide images (WSI) is still challenging mainly due to two aspects. (1) The resolution of the image is too large. (2) lacking labeled training data. Whole-slide images generally stored in a multi-resolution structure with multiple downsampled tiles. It is difficult to feed the whole image into memory without compression. Moreover, labeling images for the pathologists are time-consuming and expensive. In this paper, we study the problem of detecting breast cancer metastases in the pathological image on patch level. To address the abovementioned challenges, we propose a few-shot learning method to classify whether an image patch contains tumor cells. Specifically, we propose a patch-level unsupervised cell ranking approach, which only relies on images with limited labels. The main idea of the proposed method is that when cropping a patch A from the WSI and further cropping a sub-patch B from A, the cell number of A is always larger than that of B. Based on this observation, we make use of the unlabeled images to learn the ranking information of cell counting to extract the abstract features. Experimental results show that our method is effective to improve the patch-level classification accuracy, compared to the traditional supervised method. The source code is publicly available at https://github.com/fewshot-camelyon.

Existing GAN-based multi-view face synthesis methods rely heavily on "creating" faces, and thus they struggle in reproducing the faithful facial texture and fail to preserve identity when undergoing a large angle rotation. In this paper, we combat this problem by dividing the challenging large-angle face synthesis into a series of easy small-angle rotations, and each of them is guided by a face flow to maintain faithful facial details. In particular, we propose a Face Flow-guided Generative Adversarial Network (FFlowGAN) that is specifically trained for small-angle synthesis. The proposed network consists of two modules, a face flow module that aims to compute a dense correspondence between the input and target faces. It provides strong guidance to the second module, face synthesis module, for emphasizing salient facial texture. We apply FFlowGAN multiple times to progressively synthesize different views, and therefore facial features can be propagated to the target view from the very beginning. All these multiple executions are cascaded and trained end-to-end with a unified back-propagation, and thus we ensure each intermediate step contributes to the final result. Extensive experiments demonstrate the proposed divide-and-conquer strategy is effective, and our method outperforms the state-of-the-art on four benchmark datasets qualitatively and quantitatively.

Existing discriminative learning methods for image denoising use either a single residual learning or a nonresidual learning design. However, we observe that these two schemes perform differently with the same noise level, and yet, there have been no explorations regarding whether residual or nonresidual designs are better suited for denoising. Additionally, many discriminative denoisers are designed to learn a model that corresponds to a fixed noise level, which means that multiple models are required to recover corrupted images with noise at different levels. In this paper, we propose a dynamic dual learning network for blind image denoising, namely, DualBDNet. Instead of modeling a sole task prediction network, the proposed DualBDNet investigates the inherent relations between the residual estimation and the nonresidual estimation. In particular, DualBDNet produces task-dependent feature maps, and each part of the features is devoted to one specific task (residual/nonresidual mapping). To address different noise levels with a single network or even cases where the statistics of noise are unknown, we further introduce an embedded subnetwork into DualBDNet. One output of the subnetwork is the learning of a dynamic compositional attention to highlight the more significant task-dependent feature maps, adaptively coinciding with the extent of corruption. The other output is the learning of a weight used for fusion of the results to ensure an end-to-end manner. Extensive experiments demonstrate that the proposed DualBDNet outperforms the state-of-the-art methods on both synthetic and real noisy images without estimating the noise levels as input.

With the explosive growth of action categories, zero-shot action recognition aims to extend a well-trained model to novel/unseen classes. To bridge the large knowledge gap between seen and unseen classes, in this brief, we visually associate unseen actions with seen categories in a visually connected graph, and the knowledge is then transferred from the visual features space to semantic space via the grouped attention graph convolutional networks (GAGCNs). In particular, we extract visual features for all the actions, and a visually connected graph is built to attach seen actions to visually similar unseen categories. Moreover, the proposed grouped attention mechanism exploits the hierarchical knowledge in the graph so that the GAGCN enables propagating the visual-semantic connections from seen actions to unseen ones. We extensively evaluate the proposed method on three data sets: HMDB51, UCF101, and NTU RGB + D. Experimental results show that the GAGCN outperforms state-of-the-art methods.

Color dimensionality reduction is believed as a non-invertible process, as re-colorization results in perceptually noticeable and unrecoverable distortion. In this article, we propose to convert a color image into a grayscale image that can fully recover its original colors, and more importantly, the encoded information is discriminative and sparse, which saves storage capacity. Particularly, we design an invertible deep neural network for color encoding and decoding purposes. This network learns to generate a residual image that encodes color information, and it is then combined with a base grayscale image for color recovering. In this way, the non-differentiable compression process (e.g., JPEG) of the base grayscale image can be integrated into the network in an end-to-end manner. To further reduce the size of the residual image, we present a specific layer to enhance Sparsity Enforcing Priors (SEP), thus leading to negligible storage space. The proposed method allows color embedding on a sparse residual image while keeping a high, 35dB PSNR on average. Extensive experiments demonstrate that the proposed method outperforms state-of-the-arts in terms of image quality and tolerability to compression.

Image matting is an ill-posed problem that usually requires additional user input, such as trimaps or scribbles. Drawing a fine trimap requires a large amount of user effort, while using scribbles can hardly obtain satisfactory alpha mattes for non-professional users. Some recent deep learning–based matting networks rely on large-scale composite datasets for training to improve performance, resulting in the occasional appearance of obvious artifacts when processing natural images. In this article, we explore the intrinsic relationship between user input and alpha mattes and strike a balance between user effort and the quality of alpha mattes. In particular, we propose an interactive framework, referred to as smart scribbles, to guide users to draw few scribbles on the input images to produce high-quality alpha mattes. It first infers the most informative regions of an image for drawing scribbles to indicate different categories (foreground, background, or unknown) and then spreads these scribbles (i.e., the category labels) to the rest of the image via our well-designed two-phase propagation. Both neighboring low-level affinities and high-level semantic features are considered during the propagation process. Our method can be optimized without large-scale matting datasets and exhibits more universality in real situations. Extensive experiments demonstrate that smart scribbles can produce more accurate alpha mattes with reduced additional input, compared to the state-of-the-art matting methods.

Photorealistic multi-view face synthesis from a single image is an important but challenging problem. Existing methods mainly learn a texture mapping model from the source face to the target face. However, they fail to consider the internal deformation caused by the change of poses, leading to the unsatisfactory synthesized results for large pose variations. In this paper, we propose a Gated Deformable Face Synthesis Network to model the deformation of faces that aids the synthesis of the target face image. Specifically, we propose a dual network that consists of two modules. The first module estimates the deformation of two views in the form of convolution offsets according to the input and target poses. The second one, on the other hand, leverages the predicted deformation offsets to create the target face image. In this way, pose changes are explicitly modeled in the face generator to cope with geometric transformation, by adaptively focusing on pertinent regions of the source image. To compensate offset estimation errors, we introduce a soft-gating mechanism that enables adaptive fusion between deformable features and primitive features. Extensive experimental results on five widely-used benchmarks show that our approach performs favorably against the state-of-the-arts on multi-view face synthesis, especially for large pose changes.

Glass is very common in our daily life. Existing computer vision systems neglect it and thus may have severe consequences, e.g., a robot may crash into a glass wall. However, sensing the presence of glass is not straightforward. The key challenge is that arbitrary objects/scenes can appear behind the glass, and the content within the glass region is typically similar to those behind it. In this paper, we propose an important problem of detecting glass from a single RGB image. To address this problem, we construct a large-scale glass detection dataset (GDD) and design a glass detection network, called GDNet, which explores abundant contextual cues for robust glass detection with a novel large-field contextual feature integration (LCFI) module. Extensive experiments demonstrate that the proposed method achieves more superior glass detection results on our GDD test set than state-of-the-art methods fine-tuned for glass detection.

Temporal repetition counting aims to estimate the number of cycles of a given repetitive action. Existing deep learning methods assume repetitive actions are performed in a fixed time-scale, which is invalid for the complex repetitive actions in real life. In this paper, we tailor a context-aware and scale-insensitive framework, to tackle the challenges in repetition counting caused by the unknown and diverse cycle-lengths. Our approach combines two key insights: (1) Cycle lengths from different actions are unpredictable that require large-scale searching, but, once a coarse cycle length is determined, the variety between repetitions can be overcome by regression. (2) Determining the cycle length cannot only rely on a short fragment of video but a contextual understanding. The first point is implemented by a coarse-to-fine cycle refinement method. It avoids the heavy computation of exhaustively searching all the cycle lengths in the video, and, instead, it propagates the coarse prediction for further refinement in a hierarchical manner. We secondly propose a bidirectional cycle length estimation method for a context-aware prediction. It is a regression network that takes two consecutive coarse cycles as input, and predicts the locations of the previous and next repetitive cycles. To benefit the training and evaluation of temporal repetition counting area, we construct a new and largest benchmark, which contains 526 videos with diverse repetitive actions. Extensive experiments show that the proposed network trained on a single dataset outperforms state-of-the-art methods on several benchmarks, indicating that the proposed framework is general enough to capture repetition patterns across domains. Code and data are available in https://github.com/Xiaodomgdomg/Deep-Temporal-Repetition-Counting.

Face synthesis is an interesting yet challenging task in computer vision. It is even much harder to generate a portrait video than a single image. In this paper, we propose a novel video generation framework for synthesizing arbitrary-length face videos without any face exemplar or landmark. To overcome the synthesis ambiguity of face video, we propose a divide-and-conquer strategy to separately address the video face synthesis problem from two aspects, face identity synthesis and rearrangement. To this end, we design a cascaded network which contains three components, Identity-aware GAN (IA-GAN), Face Coherence Network, and Interpolation Network. IA-GAN is proposed to synthesize photorealistic faces with the same identity from a set of noises. Face Coherence Network is designed to re-arrange the faces generated by IA-GAN while keeping the inter-frame coherence. Interpolation Network is introduced to eliminate the discontinuity between two adjacent frames and improve the smoothness of the face video. Experimental results demonstrate that our proposed network is able to generate face video with high visual quality while preserving the identity. Statistics show that our method outperforms state-of-the-art unconditional face video generative models in multiple challenging datasets.

Abstract Cartoon is highly abstracted with clear edges, which makes it unique from the other art forms. In this paper, we focus on the essential cartoon factors of abstraction and edges, aiming to cartoonize real‐world photographs like an artist. To this end, we propose a two‐stage network, each stage explicitly targets at producing abstracted shading and crisp edges respectively. In the first abstraction stage, we propose a novel unsupervised bilateral flattening loss, which allows generating high‐quality smoothing results in a label‐free manner. Together with two other semantic‐aware losses, the abstraction stage imposes different forms of regularization for creating cartoon‐like flattened images. In the second stage we draw lines on the structural edges of the flattened cartoon with the fully supervised line drawing objective and unsupervised edge augmenting loss. We collect a cartoon‐line dataset with line tracing, and it serves as the starting point for preparing abstraction and line drawing data. We have evaluated the proposed method on a large number of photographs, by converting them to three different cartoon styles. Our method substantially outperforms state‐of‐the‐art methods in terms of visual quality quantitatively and qualitatively.

Abstract We propose a novel deep example‐based image colourization method called dense encoding pyramid network. In our study, we define the colourization as a multinomial classification problem. Given a greyscale image and a reference image, the proposed network leverages large‐scale data and then predicts colours by analysing the colour distribution of the reference image. We design the network as a pyramid structure in order to exploit the inherent multi‐scale, pyramidal hierarchy of colour representations. Between two adjacent levels, we propose a hierarchical decoder–encoder filter to pass the colour distributions from the lower level to higher level in order to take both semantic information and fine details into account during the colourization process. Within the network, a novel parallel residual dense block is proposed to effectively extract the local–global context of the colour representations by widening the network. Several experiments, as well as a user study, are conducted to evaluate the performance of our network against state‐of‐the‐art colourization methods. Experimental results show that our network is able to generate colourful, semantically correct and visually pleasant colour images. In addition, unlike fully automatic colourization that produces fixed colour images, the reference image of our network is flexible; both natural images and simple colour palettes can be used to guide the colourization.

Unsupervised domain adaptation aims to generalize a model from the label-rich source domain to the unlabeled target domain. Existing works mainly focus on aligning the global distribution statistics between source and target domains. However, they neglect distractions from the unexpected noisy samples in domain distribution estimation, leading to domain misalignment or even negative transfer. In this paper, we present an importance sampling method for domain adaptation (ISDA), to measure sample contributions according to their “informative” levels. In particular, informative samples, as well as outliers, can be effectively modeled using feature-norm and prediction entropy of the network. The importance of information is further formulated as the importance sampling losses in features and label spaces. In this way, the proposed model mitigates the noisy outliers while enhancing the important samples during domain alignment. In addition, our model is easy to implement yet effective, and it does not introduce any extra parameters. Extensive experiments on several benchmark datasets show that our method outperforms state-of-the-art methods under both the standard and partial domain adaptation settings.

In this paper, we propose a deep CNN to tackle the image restoration problem by learning formatted information. Previous deep learning based methods directly learn the mapping from corrupted images to clean images, and may suffer from the gradient exploding/vanishing problems of deep neural networks. We propose to address the image restoration problem by learning the structured details and recovering the latent clean image together, from the shared information between the corrupted image and the latent image. In addition, instead of learning the pure difference (corruption), we propose to add a residual formatting layer and an adversarial block to format the information to structured one, which allows the network to converge faster and boosts the performance. Furthermore, we propose a cross-level loss net to ensure both pixel-level accuracy and semantic-level visual quality. Evaluations on public datasets show that the proposed method performs favorably against existing approaches quantitatively and qualitatively.

Crowd counting is challenging due to unconstrained imaging factors, e.g., background clutters, non-uniform distribution of people, large scale and perspective variations. Dealing with these problems using deep neural networks requires rich prior knowledge and multi-scale contextual representations. In this paper, we propose a Cross-stage Refinement Network (CRNet) that can refine predicted density maps progressively based on hierarchical multi-level density priors. In particular, CRNet is composed of several fully convolutional networks. They are stacked together recursively with the previous output as the next input, and each of them serves to utilize previous density output to gradually correct prediction errors of crowd areas and refine the predicted density maps at different stages. Cross-stage multi-level density priors are further exploited in our recurrent framework by the cross-stage skip layers based on ConvLSTM. To cope with different challenges of unconstrained crowd scenes, we explore different crowd-specific data augmentation methods to mimic real-world scenarios and enrich crowd feature representations from different aspects. Extensive experiments show the proposed method achieves superior performances against state-of-the-art methods on four widely-used challenging benchmarks in terms of counting accuracy and density map quality. Code and models are available at this https://github.com/lytgftyf/Crowd-Counting-via-Cross-stage-Refinement-Networks.

Mobile devices usually mount a depth sensor to resolve ill-posed problems, like salient object detection on cluttered background. The main barrier of exploring RGBD data is to handle the information from two different modalities. To cope with this problem, in this paper, we propose a boundary-aware cross-modal fusion network for RGBD salient object detection. In particular, to enhance the fusion of color and depth features, we present a cross-modal feature sampling module to balance the contribution of the RGB and depth features based on the statistics of their channel values. In addition, in our multi-scale dense fusion network architecture, we not only incorporate edge-sensitive losses to preserve the boundary of the detected salient region, but also refine its structure by merging the estimated saliency maps of different scales. We accomplish the multi-scale saliency map merging using two alternative methods which produce refined saliency maps via per-pixel weighted combination and an encoder-decoder network. Extensive experimental evaluations demonstrate that our proposed framework can achieve the state-of-the-art performance on several public RGBD-based datasets.

Supervoxel segmentation algorithm has been applied as a preprocessing step for many vision tasks. However, existing supervoxel segmentation algorithms cannot generate hierarchical supervoxel segmentation well preserving the spatiotemporal boundaries in real time, which prevents the downstream applications from accurate and efficient processing. In this paper, we propose a real-time hierarchical supervoxel segmentation algorithm based on the minimum spanning tree (MST), which achieves state-of-the-art accuracy meanwhile at least 11× faster than existing methods. In particular, we present a dynamic graph updating operation into the iterative construction process of the MST, which can geometrically decrease the numbers of vertices and edges. In this way, the proposed method is able to generate arbitrary scales of supervoxels on the fly. We prove the efficiency of our algorithm that can produce hierarchical supervoxels in the time complexity of O(n) , where n denotes the number of voxels in the input video. Quantitative and qualitative evaluations on public benchmarks demonstrate that our proposed algorithm significantly outperforms the state-of-the-art algorithms in terms of supervoxel segmentation accuracy and computational efficiency. Furthermore, we demonstrate the effectiveness of the proposed method on a downstream application of video object segmentation.

Existing video salient object detection (VSOD) methods focus on exploring either short-term or long-term temporal information. However, temporal information is exploited in a global frame-level or regular grid structure, neglecting interframe structural dependencies. In this paper, we propose to learn long-term structural dependencies with a structure-evolving graph convolutional network (GCN). Particularly, we construct a graph for the entire video using a fast supervoxel segmentation method, in which each node is connected according to spatio-temporal structural similarity. We infer the inter-frame structural dependencies of salient object using convolutional operations on the graph. To prune redundant connections in the graph and better adapt to the moving salient object, we present an adaptive graph pooling to evolve the structure of the graph by dynamically merging similar nodes, learning better hierarchical representations of the graph. Experiments on six public datasets show that our method outperforms all other state-of-the-art methods. Furthermore, We also demonstrate that our proposed adaptive graph pooling can effectively improve the supervoxel algorithm in the term of segmentation accuracy.

Taking photos through a glass window leads to glare or reflection, which might distract the viewer from the scene behind the window. In this paper, we involve user interaction to tackle the ill-posedness of the reflection removal problem. Users are allowed to draw strokes or lassos to indicate the background and reflection layers. Instead of designing hand-crafted features, we propose the edge-aware cascaded networks for reflection removal. The proposed network is a two-stage pipeline. The first stage takes the edge hints converted from user guidance and the image with reflection as input, and then separates the input image into the background and reflection layers. The second stage involves a refinement network to recover the missing details of the background layers. We simulate different types of user guidance, and the networks are trained on simulated data. The cascaded networks are end-to-end and perform with a single feed-forward pass, enabling fast editing. Extensive experimental evaluations demonstrate that the proposed used-guided reflection removal network yields better performance than the state-of-the-art methods on real-world scenarios. Furthermore, we show that novice users can easily generate reflection-free images, and large improvements in reflection removal quality can be obtained in just one minute.

Top-down, goal-driven visual saliency exerts a huge influence on the human visual system for performing visual tasks. Text generations, like visual question answering (VQA) and visual question generation (VQG), have intrinsic connections with top-down saliency, which is usually involved in both VQA and VQG processes in an unsupervised manner. However, it is shown that the regions that humans choose to look at to answer questions are very different from the unsupervised attention models. In this brief, we aim to explore the intrinsic relationship between top-down saliency and text generations, and to figure out whether an accurate saliency response benefits text generation. To this end, we propose a dual supervised network with dynamic parameter prediction. Dual-supervision explicitly exploits the probabilistic correlation between the primal task top-down saliency detection and the dual task text generation, while dynamic parameter prediction encodes the given text (i.e., question or answer) into the fully convolutional network. Extensive experiments show the proposed top-down saliency method achieves the best correlation with human attention among various baselines. In addition, the proposed model can be guided by either questions or answers, and output the counterpart. Furthermore, we show that combining human-like visual question-saliency improves the performance of both answer and question generations.

Curvilinear structures are frequently observed in various images in different forms, such as blood vessels or neuronal boundaries in biomedical images. In this paper, we propose a novel curvilinear structure segmentation approach using context-aware spatio-recurrent networks. Instead of directly segmenting the whole image or densely segmenting fixed-sized local patches, our method recurrently samples patches with varied scales from the target image with learned policy and processes them locally, which is similar to the behavior of changing retinal fixations in the human visual system and it is beneficial for capturing the multi-scale or hierarchical modality of the complex curvilinear structures. In specific, the policy of choosing local patches is attentively learned based on the contextual information of the image and the historical sampling experience. In this way, with more patches sampled and refined, the segmentation of the whole image can be progressively improved. To validate our approach, comparison experiments on different types of image data are conducted and the sampling procedures for exemplar images are illustrated. We demonstrate that our method achieves the state-of-the-art performance in public datasets.

We address the problem of video representation learning without human-annotated labels. While previous efforts address the problem by designing novel self-supervised tasks using video data, the learned features are merely on a frame-by-frame basis, which are not applicable to many video analytic tasks where spatio-temporal features are prevailing. In this paper we propose a novel self-supervised approach to learn spatio-temporal features for video representation. Inspired by the success of two-stream approaches in video classification, we propose to learn visual features by regressing both motion and appearance statistics along spatial and temporal dimensions, given only the input video data. Specifically, we extract statistical concepts (fast-motion region and the corresponding dominant direction, spatio-temporal color diversity, dominant color, etc.) from simple patterns in both spatial and temporal domains. Unlike prior puzzles that are even hard for humans to solve, the proposed approach is consistent with human inherent visual habits and therefore easy to answer. We conduct extensive experiments with C3D to validate the effectiveness of our proposed approach. The experiments show that our approach can significantly improve the performance of C3D when applied to video classification tasks. Code is available at https://github.com/laura-wang/video_repres_mas.

Due to the lack of paired data, the training of image reflection removal relies heavily on synthesizing reflection images. However, existing methods model reflection as a linear combination model, which cannot fully simulate the real-world scenarios. In this paper, we inject non-linearity into reflection removal from two aspects. First, instead of synthesizing reflection with a fixed combination factor or kernel, we propose to synthesize reflection images by predicting a non-linear alpha blending mask. This enables a free combination of different blurry kernels, leading to a controllable and diverse reflection synthesis. Second, we design a cascaded network for reflection removal with three tasks: predicting the transmission layer, reflection layer, and the non-linear alpha blending mask. The former two tasks are the fundamental outputs, while the latter one being the side output of the network. This side output, on the other hand, making the training a closed loop, so that the separated transmission and reflection layers can be recombined together for training with a reconstruction loss. Extensive quantitative and qualitative experiments demonstrate the proposed synthesis and removal approaches outperforms state-of-the-art methods on two standard benchmarks, as well as in real-world scenarios.

In this paper, we propose a novel iterative multi-task framework to complete the segmentation mask of an occluded vehicle and recover the appearance of its invisible parts. In particular, firstly, to improve the quality of the segmentation completion, we present two coupled discriminators that introduce an auxiliary 3D model pool for sampling authentic silhouettes as adversarial samples. In addition, we propose a two-path structure with a shared network to enhance the appearance recovery capability. By iteratively performing the segmentation completion and the appearance recovery, the results will be progressively refined. To evaluate our method, we present a dataset, Occluded Vehicle dataset, containing synthetic and real-world occluded vehicle images. Based on this dataset, we conduct comparison experiments and demonstrate that our model outperforms the state-of-the-arts in both tasks of recovering segmentation mask and appearance for occluded vehicles. Moreover, we also demonstrate that our appearance recovery approach can benefit the occluded vehicle tracking in real-world videos.

Effectively utilizing the common information in a set of video frames is a vital aspect in video segmentation. However, existing methods that transport the common information from a prior frame to the current frame do not make use of the common information effectively. In order to address this issue, we apply a new strategy that jointly segments object through a convolutional neural network (CNN) to build a proposal-driven framework for exploiting the common information between two video frames by processing two video frames simultaneously in this letter. Moreover, proposals from the video frames are found useful for refining the segmentation results through fusing their segmentation results with the ones of the video frames. In our framework, proposals with features are generated by a faster region-CNN, and the L2 loss function is used to establish proposal pairs among proposals from the two selected frames. A new trained ResNet then keeps proposal pairs, which contain the same content, and the PSPNet model for segmentation is utilized to generate the segmentation results belonging to the frames and proposals. Finally, the proposals' segmentation results are refined using the video frames' segmentation results. The VOT 2016 segmentation dataset, the DAVIS 2017 dataset, and the SegTrack v2 dataset were used for training and testing our framework. Experimental results show that our proposal-driven segmentation framework is able to achieve higher accuracies in video segmentation challenge compared to those of the existing video segmentation methods.

Object proposal algorithms have been demonstrated to be very successful in accelerating object detection process. High object localization quality and detection recall can be obtained using thousands of proposals. However, the performance with a small number of proposals is still unsatisfactory. This paper demonstrates that the performance of a few proposals can be significantly improved with the minimal human interaction-a single touch point. To this end, we first generate hierarchical superpixels using an efficient tree-organized structure as our initial object proposals, and then select only a few proposals from them by learning an effective Convolutional neural network for objectness ranking. We explore and design an architecture to integrate human interaction with the global information of the whole image for objectness scoring, which is able to significantly improve the performance with a minimum number of object proposals. Extensive experiments show the proposed method outperforms all the state-of-the-art methods for locating the meaningful object with the touch point constraint. Furthermore, the proposed method is extended for video. By combining with the novel interactive motion segmentation cue for generating hierarchical superpixels, the performance on a single proposal is satisfactory and can be used in the interactive vision systems, such as selecting the input of a real-time tracking system.

The tracking-by-detection framework receives growing attention through the integration with the convolutional neural networks (CNNs). Existing tracking-by-detection-based methods, however, fail to track objects with severe appearance variations. This is because the traditional convolutional operation is performed on fixed grids, and thus may not be able to find the correct response while the object is changing pose or under varying environmental conditions. In this paper, we propose a deformable convolution layer to enrich the target appearance representations in the tracking-by-detection framework. We aim to capture the target appearance variations via deformable convolution, which adaptively enhances its original features. In addition, we also propose a gated fusion scheme to control how the variations captured by the deformable convolution affect the original appearance. The enriched feature representation through deformable convolution facilitates the discrimination of the CNN classifier on the target object and background. The extensive experiments on the standard benchmarks show that the proposed tracker performs favorably against the state-of-the-art methods.

Vision-based vehicle detection approaches achieve incredible success in recent years with the development of deep convolutional neural network (CNN). However, existing CNN-based algorithms suffer from the problem that the convolutional features are scale-sensitive in object detection task but it is common that traffic images and videos contain vehicles with a large variance of scales. In this paper, we delve into the source of scale sensitivity, and reveal two key issues: 1) existing RoI pooling destroys the structure of small scale objects and 2) the large intra-class distance for a large variance of scales exceeds the representation capability of a single network. Based on these findings, we present a scale-insensitive convolutional neural network (SINet) for fast detecting vehicles with a large variance of scales. First, we present a context-aware RoI pooling to maintain the contextual information and original structure of small scale objects. Second, we present a multi-branch decision network to minimize the intra-class distance of features. These lightweight techniques bring zero extra time complexity but prominent detection accuracy improvement. The proposed techniques can be equipped with any deep network architectures and keep them trained end-to-end. Our SINet achieves state-of-the-art performance in terms of accuracy and speed (up to 37 FPS) on the KITTI benchmark and a new highway dataset, which contains a large variance of scales and extremely small objects.

Image matting is an ill-posed problem. It requires a user input trimap or some strokes to obtain an alpha matte of the foreground object. A fine user input is essential to obtain a good result, which is either time consuming or suitable for experienced users who know where to place the strokes. In this paper, we explore the intrinsic relationship between the user input and the matting algorithm to address the problem of where and when the user should provide the input. Our aim is to discover the most informative sequence of regions for user input in order to produce a good alpha matte with minimum labeling efforts. To this end, we propose an active matting method with recurrent reinforcement learning. The proposed framework involves human in the loop by sequentially detecting informative regions for trivial human judgement. Comparing to traditional matting algorithms, the proposed framework requires much less efforts, and can produce satisfactory results with just 10 regions. Through extensive experiments, we show that the proposed model reduces user efforts significantly and achieves comparable performance to dense trimaps in a user-friendly manner. We further show that the learned informative knowledge can be generalized across different matting algorithms.

We present an approach to localize generic actions in egocentric videos, called temporal action proposals (TAPs), for accelerating the action recognition step. An egocentric TAP refers to a sequence of frames that may contain a generic action performed by the wearer of a head-mounted camera, e.g., taking a knife, spreading jam, pouring milk, or cutting carrots. Inspired by object proposals, this paper aims at generating a small number of TAPs, thereby replacing the popular sliding window strategy, for localizing all action events in the input video. To this end, we first propose to temporally segment the input video into action atoms, which are the smallest units that may contain an action. We then apply a hierarchical clustering algorithm with several egocentric cues to generate TAPs. Finally, we propose two actionness networks to score the likelihood of each TAP containing an action. The top ranked candidates are returned as output TAPs. Experimental results show that the proposed TAP detection framework performs significantly better than relevant approaches for egocentric action detection.

In this paper, we propose a novel -regularized optimization framework for image downscaling. The optimization is driven by two -regularized priors. The first prior, gradient-ratio prior, is based on the observation that the number of edges in the downscaled image is approximately inverse square proportional to the downscaling factor. By introducing norm sparsity to the gradient ratio, the downscaled image is able to preserve the most salient edges as well as the visual perception of the original image. The second prior, downsampling prior, is to constrain the downsampling matrix so that pixels of the downscaled image are estimated according to those optimal neighboring pixels. Extensive experiments on the Urban100 and BSDS500 data sets show that the proposed algorithm achieves superior performance over the state-of-the-arts, in terms of both quality and robustness.

In this paper, we present a novel unsupervised learning method for pixelization. Due to the difficulty in creating pixel art, preparing the paired training data for supervised learning is impractical. Instead, we propose an unsupervised learning framework to circumvent such difficulty. We leverage the dual nature of the pixelization and depixelization, and model these two tasks in the same network in a bi-directional manner with the input itself as training supervision. These two tasks are modeled as a cascaded network which consists of three stages for different purposes. GridNet transfers the input image into multi-scale grid-structured images with different aliasing effects. PixelNet associated with GridNet to synthesize pixel arts with sharp edges and perceptually optimal local structures. DepixelNet connects the previous network and aims to recover the pixelized result to the original image. For the sake of unsupervised learning, the mirror loss is proposed to hold the reversibility of feature representations in the process. In addition, adversarial, L1, and gradient losses are involved in the network to obtain pixel arts by retaining color correctness and smoothness. We show that our technique can synthesize crisper and perceptually more appropriate pixel arts than state-of-the-art image downscaling methods. We evaluate the proposed method with extensive experiments on many images. The proposed method outperforms state-of-the-art methods in terms of visual quality and user preference.

Egocentric videos, which mainly record the activities carried out by the users of wearable cameras, have drawn much research attention in recent years. Due to its lengthy content, a large number of ego-related applications have been developed to abstract the captured videos. As the users are accustomed to interacting with the target objects using their own hands, while their hands usually appear within their visual fields during the interaction, an egocentric hand detection step is involved in tasks like gesture recognition, action recognition, and social interaction understanding. In this work, we propose a dynamic region-growing approach for hand region detection in egocentric videos, by jointly considering hand-related motion and egocentric cues. We first determine seed regions that most likely belong to the hand, by analyzing the motion patterns across successive frames. The hand regions can then be located by extending from the seed regions, according to the scores computed for the adjacent superpixels. These scores are derived from four egocentric cues: contrast, location, position consistency, and appearance continuity. We discuss how to apply the proposed method in real-life scenarios, where multiple hands irregularly appear and disappear from the videos. Experimental results on public datasets show that the proposed method achieves superior performance compared with the state-of-the-art methods, especially in complicated scenarios.

Shadow removal is a challenging task as it requires the detection/annotation of shadows as well as semantic understanding of the scene. In this paper, we propose an automatic and end-to-end deep neural network (DeshadowNet) to tackle these problems in a unified manner. DeshadowNet is designed with a multi-context architecture, where the output shadow matte is predicted by embedding information from three different perspectives. The first global network extracts shadow features from a global view. Two levels of features are derived from the global network and transferred to two parallel networks. While one extracts the appearance of the input image, the other one involves semantic understanding for final prediction. These two complementary networks generate multi-context features to obtain the shadow matte with fine local details. To evaluate the performance of the proposed method, we construct the first large scale benchmark with 3088 image pairs. Extensive experiments on two publicly available benchmarks and our large-scale benchmark show that the proposed method performs favorably against several state-of-the-art methods.

Subitizing (i.e., instant judgement on the number) and detection of salient objects are human inborn abilities. These two tasks influence each other in the human visual system. In this paper, we delve into the complementarity of these two tasks. We propose a multi-task deep neural network with weight prediction for salient object detection, where the parameters of an adaptive weight layer are dynamically determined by an auxiliary subitizing network. The numerical representation of salient objects is therefore embedded into the spatial representation. The proposed joint network can be trained end-to-end using backpropagation. Experiments show the proposed multi-task network outperforms existing multi-task architectures, and the auxiliary subitizing network provides strong guidance to salient object detection by reducing false positives and producing coherent saliency maps. Moreover, the proposed method is an unconstrained method able to handle images with/without salient objects. Finally, we show state-of-the-art performance on different salient object datasets.

Image captioning has recently received much attention. Existing approaches, however, are limited to describing images with simple contextual information, which typically generate one sentence to describe each image with only a single contextual emphasis. In this paper, we address this limitation from a user perspective with a novel approach. Given some keywords as additional inputs, the proposed method would generate various descriptions according to the provided guidance. Hence, descriptions with different focuses can be generated for the same image. Our method is based on a new Context-dependent Bilateral Long Short-Term Memory (CDB-LSTM) model to predict a keyword-driven sentence by considering the word dependence. The word dependence is explored externally with a bilateral pipeline, and internally with a unified and joint training process. Experiments on the MS COCO dataset demonstrate that the proposed approach not only significantly outperforms the baseline method but also shows good adaptation and consistency with various keywords.

We propose a two-stage method for face hallucination. First, we generate facial components of the input image using CNNs. These components represent the basic facial structures. Second, we synthesize fine-grained facial structures from high resolution training images. The details of these structures are transferred into facial components for enhancement. Therefore, we generate facial components to approximate ground truth global appearance in the first stage and enhance them through recovering details in the second stage. The experiments demonstrate that our method performs favorably against state-of-the-art methods.

This paper presents a novel locality sensitive histogram (LSH) algorithm for visual tracking. Unlike the conventional image histogram that counts the frequency of occurrence of each intensity value by adding ones to the corresponding bin, an LSH is computed at each pixel location, and a floating-point value is added to the corresponding bin for each occurrence of an intensity value. The floating-point value exponentially reduces with respect to the distance to the pixel location where the histogram is computed. An efficient algorithm is proposed that enables the LSHs to be computed in time linear in the image size and the number of bins. In addition, this efficient algorithm can be extended to exploit color images. A robust tracking framework based on the LSHs is proposed, which consists of two main components: a new feature for tracking that is robust to illumination change and a novel multiregion tracking algorithm that runs in real time even with hundreds of regions. Extensive experiments demonstrate that the proposed tracking framework outperforms the state-of-the-art methods in challenging scenarios, especially when the illumination changes dramatically. Evaluation using the latest benchmark shows that our algorithm is the top performer.

Object proposal detection is an effective way of accelerating object recognition. Existing proposal methods are mostly based on detecting object boundaries, which may not be effective for cluttered backgrounds. In this paper, we leverage stereopsis as a robust and effective solution for generating object proposals. We first obtain a set of candidate bounding boxes through adaptive transformation, which fits the bounding boxes tightly to object boundaries detected by rough depth and color information. A two-level hierarchy composed of proposal and cluster levels is then constructed to estimate object locations in an efficient and accurate manner. Three stereo-based cues "exactness," "focus," and "distribution" are proposed for objectness estimation. Two-level hierarchical ranking is proposed to accurately obtain ranked object proposals. A stereo data set with 400 labeled stereo image pairs is constructed to evaluate the performance of the proposed method in both indoor and outdoor scenes. Extensive experimental evaluations show that the proposed stereo-based approach achieves a better performance than the state of the arts with either a small or a large number of object proposals. As stereopsis can be a complement to the color information, the proposed method can be integrated with existing proposal methods to obtain superior results.

Numerous efforts have been made to design various low-level saliency cues for RGBD saliency detection, such as color and depth contrast features as well as background and color compactness priors. However, how these low-level saliency cues interact with each other and how they can be effectively incorporated to generate a master saliency map remain challenging problems. In this paper, we design a new convolutional neural network (CNN) to automatically learn the interaction mechanism for RGBD salient object detection. In contrast to existing works, in which raw image pixels are fed directly to the CNN, the proposed method takes advantage of the knowledge obtained in traditional saliency detection by adopting various flexible and interpretable saliency feature vectors as inputs. This guides the CNN to learn a combination of existing features to predict saliency more effectively, which presents a less complex problem than operating on the pixels directly. We then integrate a superpixel-based Laplacian propagation framework with the trained CNN to extract a spatially consistent saliency map by exploiting the intrinsic structure of the input image. Extensive quantitative and qualitative experimental evaluations on three data sets demonstrate that the proposed method consistently outperforms the state-of-the-art methods.

Top-down saliency detection is a knowledge-driven search task. While some previous methods aim to learn this "knowledge" from category-specific data, others transfer existing annotations in a large dataset through appearance matching. In contrast, we propose in this paper a locateby-exemplar strategy. This approach is challenging, as we only use a few exemplars (up to 4) and the appearances among the query object and the exemplars can be very different. To address it, we design a two-stage deep model to learn the intra-class association between the exemplars and query objects. The first stage is for learning object-to-object association, and the second stage is to learn background discrimination. Extensive experimental evaluations show that the proposed method outperforms different baselines and the category-specific models. In addition, we explore the influence of exemplar properties, in terms of exemplar number and quality. Furthermore, we show that the learned model is a universal model and offers great generalization to unseen objects.

In this paper, we present a real-time salient object detection system based on the minimum spanning tree. Due to the fact that background regions are typically connected to the image boundaries, salient objects can be extracted by computing the distances to the boundaries. However, measuring the image boundary connectivity efficiently is a challenging problem. Existing methods either rely on superpixel representation to reduce the processing units or approximate the distance transform. Instead, we propose an exact and iteration free solution on a minimum spanning tree. The minimum spanning tree representation of an image inherently reveals the object geometry information in a scene. Meanwhile, it largely reduces the search space of shortest paths, resulting an efficient and high quality distance transform algorithm. We further introduce a boundary dissimilarity measure to compliment the shortage of distance transform for salient object detection. Extensive evaluations show that the proposed algorithm achieves the leading performance compared to the state-of-the-art methods in terms of efficiency and accuracy.

The locality sensitive histogram (LSH) injects spatial information into the local histogram in an efficient manner, and has been demonstrated to be very effective for visual tracking. In this paper, we explore the application of this efficient histogram in two important problems. We first extend the LSH to linear time bilateral filtering, and then propose a new type of histogram for efficiently computing edge-preserving nearest neighbor fields (NNFs). While the existing histogram-based bilateral filtering methods are the state of the art for efficient grayscale image processing, they are limited to box spatial filter kernels only. In our first application, we address this limitation by expressing the bilateral filter as a simple ratio of linear functions of the LSH, which is able to extend the box spatial kernel to an exponential kernel. The computational complexity of the proposed bilateral filter is linear in the number of image pixels. In our second application, we derive a new bilateral weighted histogram (BWH) for NNF. The new histogram maintains the efficiency of LSH, which allows approximate NNF to be computed independent of patch size. In addition, BWH takes both spatial and color information into account, and thus provides higher accuracy for histogram-based matching, especially around color edges.

In this paper, we propose a new approach to generate oriented object proposals (OOPs) to reduce the detection error caused by various orientations of the object. To this end, we propose to efficiently locate object regions according to pixelwise object probability, rather than measuring the objectness from a set of sampled windows. We formulate the proposal generation problem as a generative probabilistic model such that object proposals of different shapes (i.e., sizes and orientations) can be produced by locating the local maximum likelihoods. The new approach has three main advantages. First, it helps the object detector handle objects of different orientations. Second, as the shapes of the proposals may vary to fit the objects, the resulting proposals are tighter than the sampling windows with fixed sizes. Third, it avoids massive window sampling, and thereby reducing the number of proposals while maintaining a high recall. Experiments on the PASCAL VOC 2007 dataset show that the proposed OOP outperforms the state-of-the-art fast methods. Further experiments show that the rotation invariant property helps a class-specific object detector achieve better performance than the state-of-the-art proposal generation methods in either object rotation scenarios or general scenarios. Generating OOPs is very fast and takes only 0.5s per image.

Image decolorization is a fundamental problem for many real-world applications, including monochrome printing and photograph rendering. In this paper, we propose a new color-to-gray conversion method that is based on a region-based saliency model. First, we construct a parametric color-to-gray mapping function based on global color information as well as local contrast. Second, we propose a region-based saliency model that computes visual contrast among pixel regions. Third, we minimize the salience difference between the original color image and the output grayscale image in order to preserve contrast discrimination. To evaluate the performance of the proposed method in preserving contrast in complex scenarios, we have constructed a new decolorization data set with 22 images, each of which contains abundant colors and patterns. Extensive experimental evaluations on the existing and the new data sets show that the proposed method outperforms the state-of-the-art methods quantitatively and qualitatively.

This paper presents a novel locality sensitive histogram algorithm for visual tracking. Unlike the conventional image histogram that counts the frequency of occurrences of each intensity value by adding ones to the corresponding bin, a locality sensitive histogram is computed at each pixel location and a floating-point value is added to the corresponding bin for each occurrence of an intensity value. The floating-point value declines exponentially with respect to the distance to the pixel location where the histogram is computed, thus every pixel is considered but those that are far away can be neglected due to the very small weights assigned. An efficient algorithm is proposed that enables the locality sensitive histograms to be computed in time linear in the image size and the number of bins. A robust tracking framework based on the locality sensitive histograms is proposed, which consists of two main components: a new feature for tracking that is robust to illumination changes and a novel multi-region tracking algorithm that runs in real time even with hundreds of regions. Extensive experiments demonstrate that the proposed tracking framework outperforms the state-of-the-art methods in challenging scenarios, especially when the illumination changes dramatically.

Stereo images and videos are very popular in recent years, and techniques for processing this media are attracting a lot of attention. In this paper, we extend the shift-map method for stereo image editing. Our method simultaneously processes the left and right images on pixel level using a global optimization algorithm. It enforces photo consistence between the two images and preserves 3D scene structures. It also addresses the occlusion and disocclusion problem, which may enable many stereo image editing functions, such as depth mapping, object depth adjustment and non-homogeneous image resizing. Our experiments show that the proposed method produces high quality results in various editing functions.

Abstract Capturing fine details of turbulence on a coarse grid is one of the main tasks in real‐time fluid simulation. Existing methods for doing this have various limitations. In this paper, we propose a new turbulence method that uses a refined second vorticity confinement method, referred to as robust second vorticity confinement, and a synthesis scheme to create highly turbulent effects from coarse grid. The new technique is sufficiently stable to efficiently produce highly turbulent flows, while allowing intuitive control of vortical structures. Second vorticity confinement captures and defines the vortical features of turbulence on a coarse grid. However, due to the stability problem, it cannot be used to produce highly turbulent flows. In this work, we propose a robust formulation to improve the stability problem by making the positive diffusion term to vary with helicity adaptively. In addition, we also employ our new method to procedurally synthesize the high‐resolution flow fields. As shown in our results, this approach produces stable high‐resolution turbulence very efficiently.

Smoke simulation is one of the interesting topics in computer animation and it usually involves turbulence generation. Efficient generation of realistic turbulent flows becomes one of the challenges in smoke simulation. Vortex particle method, which is a hybrid method that combines grid-based and particle-based approaches, is often used for generating turbulent details. However, it may cause irrational artifacts due to its initial condition and vorticity forcing approach used. In this paper, a new vorticity forcing approach based on the spatial adaptive vorticity confinement is proposed to address this problem. In this approach, the spatial adaptive vorticity confinement force varies with helicity, leading to the fact that the grid-based simulation driven by the vortex particle is now based on the velocity field. Furthermore, we introduce an adaptive vortex particle approach to improve the computational efficiency of the simulation by making the influencing region adapt with the velocity and eliminating those particles with zero velocity in the vorticity forcing method. A parallel smoke simulator integrating our approaches has been implemented using GPUs with CUDA. Experimental results demonstrate that our proposed methods are efficient and effective for real-time smoke simulation.

Abstract Turbulence modeling has recently drawn many attentions in fluid animation to generate small‐scale rolling features. Being one of the widely adopted approaches, vorticity confinement method re‐injects lost energy dissipation back to the flow. However, previous works suffer from deficiency when large vorticity coefficientεis used, due to the fact that constantεis applied all over the simulated domain. In this paper, we propose a novel approach to enhance the visual effect by employing an adaptive vorticity confinement which varies the strength with respect to the helicity instead of a user‐defined constant. To further improve fine details in turbulent flows, we are not only applying our proposed vorticity confinement to low‐resolution grid, but also on a finer grid to generate sub‐grid level turbulence. Since the incompressible Navier–Stokes equations are solved only in low‐resolution grid, this saves a significant amount of computation. Several experiments demonstrate that our method can produce realistic smoke animation with enhanced turbulence effects in real‐time. Copyright © 2011 John Wiley & Sons, Ltd.