Searcharxiv⌕ Search

arXiv subjects

Gangshan Wu

Publications and source records attributed to Gangshan Wu.

At least 19 recordsLinked to original sources
arXiv2026

Memory Forcing: Attendable Mid-Horizon History for Streaming Video Generation

Autoregressive video diffusion enables causal video streaming without a bidirectional pass over the full clip, but existing few-step systems usually retain only the opening and most recent frames in a fixed-size KV cache. Once an event leaves this window, later frames can no longer attend to it, a failure we term mid-horizon forgetting. We present Memory Forcing, a few-step streaming method that preserves this missing history without increasing the cache size. Its Archive \& Working Banks partition the cache into sink, archive, and working regions, retaining diverse intermediate events alongside recent motion under fixed memory. Because absolute temporal indices drift outside the training range, Bank-aware RoPE reassigns indices at attention time so each bank remains distinguishable. At 1.3B, Memory Forcing leads on longer clips, shows the smallest drop from 5s to 60s among methods reporting all four lengths, and preserves subjects and scenes through leave-and-return. The same design scales to Wan2.2 5B, producing more physically plausible, realistic, and dynamic videos and, to our knowledge, the first public 5B model on this forcing line.

cs.CV↗
arXiv2026

SPLG-Mamba: Structure-Preserving Local-Global Mamba Network for Salient Object Detection in Optical Remote Sensing Images

Salient object detection in optical remote sensing images (ORSI-SOD) requires dense predictions that preserve object completeness and structural continuity under complex backgrounds, scale variation, and irregular object shapes. Existing methods often localize salient regions, but their predictions may still suffer from structural degradation, including fragmented, incomplete, or locally missing foreground responses. This degradation is closely related to hierarchical feature propagation, where shallow details can introduce texture-induced background responses, deep semantics may over-smooth weak structures, and uncontrolled cross-scale fusion can disturb coherent regions. To address this issue, we propose a novel Structure-Preserving Local-Global Mamba Network, SPLG-Mamba, for ORSI-SOD. Specifically, SPLG-Mamba integrates Smooth-Detail Recalibration (SDR), hierarchy-aware Local-Global Mamba, and Gated Cross-Scale Fusion (GCSF). SDR recalibrates smoothed responses and detail residuals before state-space modeling, Local-Global Mamba assigns local modeling to shallow feature levels and global modeling to deep feature levels, and GCSF controls cross-scale detail injection during decoding. Experiments on ORSSD, EORSSD, and ORSI-4199 demonstrate state-of-the-art results and improved structural completeness and continuity. The code is available at https://github.com/yxu9910/SPLG-Mamba

cs.CV↗
arXiv2026

VL-UniTrack: A Unified Framework with Visual-Language Prompts for UAV-Ground Visual Tracking

UAV-ground visual tracking (UGVT) aims to simultaneously track the same object from both the UAV and the ground view. However, existing two-stream methods suffer from isolated feature extraction and rely heavily on implicit appearance matching, which struggles to establish reliable correspondence under drastic view differences, leading to tracking unreliability. To address these limitations, we propose VL-UniTrack, a fully unified framework enhanced by visual-language prompts. By encoding features from both views within a single shared encoder, our method breaks the barrier of feature isolation to facilitate sufficient cross-view interaction. To overcome the ambiguity caused by relying solely on appearance matching, we design visual-language geometric prompting module, which fuses language descriptions with visual features to generate learnable prompts. These prompts are then fed into our prompt-guided cross-view adapter module to enable sufficient cross-view feature interaction and to guide the learning of view-specific feature representations. Furthermore, a confidence-modulated mutual distillation loss is proposed to regularize the training by mitigating noise propagation. Extensive experiments demonstrate that our method achieves state-of-the-art performance on the latest benchmark. The code can be downloaded in https://github.com/xuboyue1999/VL-UniTrack.git

cs.CV↗
arXiv2026

S$^3$AM: A Single-Stream SAM with Reliability-Calibrated Frequency Adapter for Multi-modal Salient Object Detection

Vision foundation models have recently advanced multi-modal salient object detection (MSOD) through parameter-efficient tuning and prompt learning. However, existing Segment Anything Model (SAM)-adapted MSOD methods often rely on dual-stream encoders or auxiliary prompt generators, leading to redundant computation. Although a single-stream alternative can reduce this cost, early fusion may also propagate noisy or misaligned auxiliary high-frequency cues through the backbone. In this paper, we propose a novel single-stream framework that integrates reliability-calibrated frequency adaptation into the adopted SAM backbone for MSOD. It avoids duplicated foundation backbones while explicitly controlling auxiliary frequency injection. Specifically, we design a mixture of frequency experts module, which uses the stationary wavelet transform to decompose each modality and aggregate cross-modal frequency information. We further introduce a reliability-calibrated frequency adapter with a dual-gate calibration mechanism, which selectively propagates the calibrated residual across transformer stages while jointly controlling its injection strength and cross-modal reliability. A hypernetwork-guided semantic-structural decoder then combines semantic mask features from the adopted backbone with Mamba-based structural detail recovery. Comprehensive experiments on RGB-D, RGB-T, and RGB-NIR salient object detection benchmarks validate that the proposed framework achieves competitive performance with only 12.20M trainable parameters, accounting for 5.4\% of the total parameters. The code will be available at https://github.com/xuboyue1999/SSSAM.

cs.CV↗
arXiv2026

Enabling Proactive Spoken Turns via a Generalized Style-Aware Full-Duplex Framework

Compared with half-duplex dialogue systems where the system waits for user turn completion before it responds, natural full-duplex dialogue systems require agents to act proactively in real time, including timely interruptions and backchannels. This creates a key challenge: improving turn timing without sacrificing response quality. To address limitations in realistic proactive turn-taking, we build a generalized style-aware full-duplex framework with three key components. Firstly, we propose LPS-TC, a Lightweight Proactive Speech Turn Controller for plug-and-play integration. It features a fine-grained action space covering both reactive and proactive turn behaviors, enabling half-duplex models with full-duplex capabilities and enhancing existing full-duplex models with superior timing control. Secondly, we construct WildTurn, a large-scale, real-world English dataset containing approximately 2,981 hours of filtered multi-turn stereo conversations from face-to-face and telephone conversations, annotated with five turn-taking and five backchanneling styles. Trained on WildTurn, LPS-TC exhibits rich spoken dynamics that are not captured by existing static full-duplex benchmarks. Thirdly, we introduce a two-tier evaluation scheme that assesses both chunk-level timing precision and turn-level interaction quality under realistic streaming constraints. Our experiments, integrating LPS-TC with half-duplex models like Qwen2.5-Omni and full-duplex models like Freeze-Omni, showcase its superior performance in timing appropriateness and response quality. Our framework also demonstrates fine-grained style controllability and strong generalizability, enabling more natural and human-like spoken interactions.

cs.CL↗
arXiv2026

VIABench: A Comprehensive Video Benchmark Collected from Blind Individuals for Visual Impairment Assistance

Visually impaired individuals (VIIs) encounter significant daily challenges due to limited access to visual information. Although Multimodal Large Language Models (MLLMs) have achieved impressive results on general vision and language tasks, their practical utility in real-world blind assistance still remains largely underexplored. To fill this gap, we introduce VIABench, a comprehensive video benchmark specifically designed to evaluate MLLMs in Visually Impaired Assistance scenarios using first-person videos recorded or shared by VIIs themselves. VIABench defines three core tasks, each targeting a distinct requirement in visual assistance. Proactive Reminder: Assesses the model's ability to interpret ongoing video content while proactively anticipating and verbally describing upcoming navigation-critical events; Visual Question Answering (VQA): Evaluates the model's capacity to answer user-posed questions about the environment or objects within the video; Vision-Guided Interaction: Tests context-aware reasoning to accomplish intentional interactions between user and environment. To ensure a robust and fair evaluation, we propose a rigorous benchmarking pipeline that supports both online (real-time) and offline settings. Our experiments demonstrate that current MLLMs still struggle to deliver comprehensive support for VIIs, especially in the Proactive Reminder task, which demands accurate anticipation and real-time responsiveness. We hope VIABench will drive future research toward developing customized MLLMs for real-world assistance, ultimately improving navigation and interaction experiences for visually impaired individuals. Code and data will be released at https://github.com/MCG-NJU/VIABench.

cs.CV↗
arXiv2026

CoMo: Learning Continuous Latent Motion from Internet Videos for Scalable Robot Learning

Unsupervised learning of latent motion from Internet videos is crucial for robot learning. Existing discrete methods generally mitigate the shortcut learning caused by extracting excessive static backgrounds through vector quantization with a small codebook size. However, they suffer from information loss and struggle to capture more complex and fine-grained dynamics. Moreover, there is an inherent gap between the distribution of discrete latent motion and continuous robot action, which hinders the joint learning of a unified policy. We propose CoMo, which aims to learn more precise continuous latent motion from internet-scale videos. CoMo employs an early temporal difference (Td) mechanism to increase the shortcut learning difficulty and explicitly enhance motion cues. Additionally, to ensure latent motion better captures meaningful foregrounds, we further propose a temporal contrastive learning (Tcl) scheme. Specifically, positive pairs are constructed with a small future frame temporal offset, while negative pairs are formed by directly reversing the temporal direction. The proposed Td and Tcl work synergistically and effectively ensure that the latent motion focuses better on the foreground and reinforces motion cues. Critically, CoMo exhibits strong zeroshot generalization, enabling it to generate effective pseudo action labels for unseen videos. Extensive simulated and real-world experiments show that policies co-trained with CoMo pseudo action labels achieve superior performance with both diffusion and auto-regressive architectures.

cs.CV↗
arXiv2026

NTIRE 2025 Challenge on Image Super-Resolution (x4): Methods and Results

This paper presents the NTIRE 2025 image super-resolution ($\times$4) challenge, one of the associated competitions of the 10th NTIRE Workshop at CVPR 2025. The challenge aims to recover high-resolution (HR) images from low-resolution (LR) counterparts generated through bicubic downsampling with a $\times$4 scaling factor. The objective is to develop effective network designs or solutions that achieve state-of-the-art SR performance. To reflect the dual objectives of image SR research, the challenge includes two sub-tracks: (1) a restoration track, emphasizes pixel-wise accuracy and ranks submissions based on PSNR; (2) a perceptual track, focuses on visual realism and ranks results by a perceptual score. A total of 286 participants registered for the competition, with 25 teams submitting valid entries. This report summarizes the challenge design, datasets, evaluation protocol, the main results, and methods of each team. The challenge serves as a benchmark to advance the state of the art and foster progress in image SR.

cs.CV↗
arXiv2026

RASR: Retrieval-Augmented Super Resolution for Practical Reference-based Image Restoration

Reference-based Super Resolution (RefSR) improves upon Single Image Super Resolution (SISR) by leveraging high-quality reference images to enhance texture fidelity and visual realism. However, a critical limitation of existing RefSR approaches is their reliance on manually curated target-reference image pairs, which severely constrains their practicality in real-world scenarios. To overcome this, we introduce Retrieval-Augmented Super Resolution (RASR), a new and practical RefSR paradigm that automatically retrieves semantically relevant high-resolution images from a reference database given only a low-quality input. This enables scalable and flexible RefSR in realistic use cases, such as enhancing mobile photos taken in environments like zoos or museums, where category-specific reference data (e.g., animals, artworks) can be readily collected or pre-curated. To facilitate research in this direction, we construct RASR-Flickr30, the first benchmark dataset designed for RASR. Unlike prior datasets with fixed target-reference pairs, RASR-Flickr30 provides per-category reference databases to support open-world retrieval. We further propose RASRNet, a strong baseline that combines a semantic reference retriever with a diffusion-based RefSR generator. It retrieves relevant references based on semantic similarity and employs a diffusion-based generator enhanced with semantic conditioning. Experiments on RASR-Flickr30 demonstrate that RASRNet consistently improves over SISR baselines, achieving +0.38 dB PSNR and -0.0131 LPIPS, while generating more realistic textures. These findings highlight retrieval augmentation as a promising direction to bridge the gap between academic RefSR research and real-world applicability.

cs.CV↗
arXiv2026

ObjFiller3D: Scaling 3D Object Inpainting to Dense Multi-View Consistency

3D object inpainting is commonly achieved via multi-view 2D image completion, yet independently inpainted views often suffer from cross-view inconsistencies, leading to blurred textures, geometric discontinuities, and visual artifacts in the reconstructed 3D objects. To overcome these limitations, we propose ObjFiller-3D, a novel method designed for the completion and editing of high-quality and consistent 3D objects. Instead of relying on sparse-view editing or per-view 2D inpainting, our method jointly optimizes a sequence of densely sampled views along a $360^\circ$ trajectory, enabling global coherence across viewpoints. We design a new framework with three complementary components: a Temporal-Driven Generative Encoder for modeling dense-view dependencies, a Semantic-Aware Completion Encoder for object-level inpainting, and a Cycle-Consistent 3D Encoder that enforces global coherence through a closed-loop formulation. Our framework also supports reference-guided 3D inpainting, allowing fine-grained control over appearance. Extensive experiments on diverse datasets demonstrate that ObjFiller-3D significantly outperforms prior methods, achieving higher reconstruction fidelity (PSNR 26.6 vs.\ 15.9 of NeRFiller) and perceptual quality (LPIPS 0.19 vs.\ 0.25 of Instant3dit), while reducing reconstruction time from over 40 minutes to under 10 minutes. These results highlight the effectiveness and practical potential of our approach for real-world 3D editing applications. Project page: https://objfiller3d.github.io/ Code: https://github.com/objfiller3d/ObjFiller-3D .

cs.CV↗
arXiv2026

SAM 2++: Tracking Anything at Any Granularity

Due to the varying granularity of target states across different tasks, most existing trackers are tailored to a single task, which specificity limits their generalization, preventing them from effectively utilizing multi-task training data and leading to redundancy in both model design and parameters. Although recent unified vision models share partial architectures across tasks, they usually retain task-specific interfaces and overlook the common tracking principle behind different granularities, leaving a gap for truly unified video tracking. To unify video tracking tasks, we present SAM 2++, a unified framework that can handle target states at different granularities, including masks, boxes, and points, through an integrated design of prompt encoding, output decoding, and memory representation. First, to handle different target granularities, we design task-specific prompts that map diverse task inputs into general prompt embeddings, together with a Unified Decoder that produces task results in a common output form without redesigning the overall pipeline. Next, to satisfy memory matching, the core operation of tracking, we introduce a task-adaptive memory mechanism that unifies memory across different granularities while preserving their distinct state semantics, preventing full parameter sharing from causing interference across granularities. Finally, we introduce Tracking-Any-Granularity, the first large and diverse video tracking dataset with rich annotations at three granularities. It is constructed through a customized data engine with phased manual annotation and model-assisted completion, providing a comprehensive resource for training, benchmarking, and analyzing unified tracking models. Comprehensive experiments confirm that SAM 2++ sets a new state of the art across diverse tracking tasks at different granularities, establishing a unified and robust tracking framework.

cs.CV↗
arXiv2026

SteadyDancer: Harmonized and Coherent Human Image Animation with First-Frame Preservation

Preserving first-frame identity while ensuring precise motion control is a fundamental challenge in human image animation. The Image-to-Motion Binding process of the dominant Reference-to-Video (R2V) paradigm overlooks critical spatio-temporal misalignments common in real-world applications, leading to failures such as identity drift and visual artifacts. We introduce SteadyDancer, an Image-to-Video (I2V) paradigm-based framework that achieves harmonized and coherent animation and is the first to ensure first-frame preservation robustly. Firstly, we propose a Condition-Reconciliation Mechanism to harmonize the two conflicting conditions, enabling precise control without sacrificing fidelity. Secondly, we design Synergistic Pose Modulation Modules to generate an adaptive and coherent pose representation that is highly compatible with the reference image. Finally, we employ a Staged Decoupled-Objective Training Pipeline that hierarchically optimizes the model for motion fidelity, visual quality, and temporal coherence. Experiments demonstrate that SteadyDancer achieves state-of-the-art performance in both appearance fidelity and motion control, while requiring significantly fewer training resources than comparable methods. The model has been publicly released at \url{https://mcg-nju.github.io/steadydancer-web}.

cs.CV↗
arXiv2026

Disentangled Textual Priors for Diffusion-based Image Super-Resolution

Image Super-Resolution (SR) aims to reconstruct high-resolution images from degraded low-resolution inputs. While diffusion-based SR methods offer powerful generative capabilities, their performance heavily depends on how semantic priors are structured and integrated into the generation process. Existing approaches often rely on entangled or coarse-grained priors that mix global layout with local details, or conflate structural and textural cues, thereby limiting semantic controllability and interpretability. In this work, we propose DTPSR, a novel diffusion-based SR framework that introduces disentangled textual priors along two complementary dimensions: spatial hierarchy (global vs. local) and frequency semantics (low- vs. high-frequency). By explicitly separating these priors, DTPSR enables the model to simultaneously capture scene-level structure and object-specific details with frequency-aware semantic guidance. The corresponding embeddings are injected via specialized cross-attention modules, forming a progressive generation pipeline that reflects the semantic granularity of visual content, from global layout to fine-grained textures. To support this paradigm, we construct DisText-SR, a large-scale dataset containing approximately 95,000 image-text pairs with carefully disentangled global, low-frequency, and high-frequency descriptions. To further enhance controllability and consistency, we adopt a multi-branch classifier-free guidance strategy with frequency-aware negative prompts to suppress hallucinations and semantic drift. Extensive experiments on synthetic and real-world benchmarks show that DTPSR achieves high perceptual quality, competitive fidelity, and strong generalization across diverse degradation scenarios.

cs.CV↗
arXiv2026

GLAD: Generative Language-Assisted Visual Tracking for Low-Semantic Templates

Vision-language tracking has gained increasing attention in many scenarios. This task simultaneously deals with visual and linguistic information to localize objects in videos. Despite its growing utility, the development of vision-language tracking methods remains in its early stage. Current vision-language trackers usually employ Transformer architectures for interactive integration of template, search, and text features. However, persistent challenges about low-semantic images including prevalent image blurriness, low resolution and so on, may compromise model performance through degraded cross-modal understanding. To solve this problem, language assistance is usually used to deal with the obstacles posed by low-semantic images. However, due to the existing gap between current textual and visual features, direct concatenation and fusion of these features may have limited effectiveness. To address these challenges, we introduce a pioneering Generative Language-AssisteD tracking model, GLAD, which utilizes diffusion models for the generative multi-modal fusion of text description and template image to bolster compatibility between language and image and enhance template image semantic information. Our approach demonstrates notable improvements over the existing fusion paradigms. Blurry and semantically ambiguous template images can be restored to improve multi-modal features in the generative fusion paradigm. Experiments show that our method establishes a new state-of-the-art on multiple benchmarks and achieves an impressive inference speed. The code and models will be released at: https://github.com/Confetti-lxy/GLAD

cs.CV↗
arXiv2026

VMonarch: Efficient Video Diffusion Transformers with Structured Attention

The quadratic complexity of the attention mechanism severely limits the context scalability of Video Diffusion Transformers (DiTs). We find that the highly sparse spatio-temporal attention patterns exhibited in Video DiTs can be naturally represented by the Monarch matrix. It is a class of structured matrices with flexible sparsity, enabling sub-quadratic attention via an alternating minimization algorithm. Accordingly, we propose VMonarch, a novel attention mechanism for Video DiTs that enables efficient computation over the dynamic sparse patterns with structured Monarch matrices. First, we adapt spatio-temporal Monarch factorization to explicitly capture the intra-frame and inter-frame correlations of the video data. Second, we introduce a recomputation strategy to mitigate artifacts arising from instabilities during alternating minimization of Monarch matrices. Third, we propose a novel online entropy algorithm fused into FlashAttention, enabling fast Monarch matrix updates for long sequences. Extensive experiments demonstrate that VMonarch achieves comparable or superior generation quality to full attention on VBench after minimal tuning. It overcomes the attention bottleneck in Video DiTs, reduces attention FLOPs by a factor of 17.5, and achieves a speedup of over 5x in attention computation for long videos, surpassing state-of-the-art sparse attention methods at 90% sparsity.

cs.CV↗
arXiv2025

SwiTrack: Tri-State Switch for Cross-Modal Object Tracking

Cross-modal object tracking (CMOT) is an emerging task that maintains target consistency while the video stream switches between different modalities, with only one modality available in each frame, mostly focusing on RGB-Near Infrared (RGB-NIR) tracking. Existing methods typically connect parallel RGB and NIR branches to a shared backbone, which limits the comprehensive extraction of distinctive modality-specific features and fails to address the issue of object drift, especially in the presence of unreliable inputs. In this paper, we propose SwiTrack, a novel state-switching framework that redefines CMOT through the deployment of three specialized streams. Specifically, RGB frames are processed by the visual encoder, while NIR frames undergo refinement via a NIR gated adapter coupled with the visual encoder to progressively calibrate shared latent space features, thereby yielding more robust cross-modal representations. For invalid modalities, a consistency trajectory prediction module leverages spatio-temporal cues to estimate target movement, ensuring robust tracking and mitigating drift. Additionally, we incorporate dynamic template reconstruction to iteratively update template features and employ a similarity alignment loss to reinforce feature consistency. Experimental results on the latest benchmarks demonstrate that our tracker achieves state-of-the-art performance, boosting precision rate and success rate gains by 7.2\% and 4.3\%, respectively, while maintaining real-time tracking at 65 frames per second. Code and models are available at https://github.com/xuboyue1999/SwiTrack.git.

cs.CV↗
arXiv2025

Towards Practical Real-Time Low-Latency Music Source Separation

In recent years, significant progress has been made in the field of deep learning for music demixing. However, there has been limited attention on real-time, low-latency music demixing, which holds potential for various applications, such as hearing aids, audio stream remixing, and live performances. Additionally, a notable tendency has emerged towards the development of larger models, limiting their applicability in certain scenarios. In this paper, we introduce a lightweight real-time low-latency model called Real-Time Single-Path TFC-TDF UNET (RT-STT), which is based on the Dual-Path TFC-TDF UNET (DTTNet). In RT-STT, we propose a feature fusion technique based on channel expansion. We also demonstrate the superiority of single-path modeling over dual-path modeling in real-time models. Moreover, we investigate the method of quantization to further reduce inference time. RT-STT exhibits superior performance with significantly fewer parameters and shorter inference times compared to state-of-the-art models.

cs.SD↗
arXiv2025

Pure-Pass: Fine-Grained, Adaptive Masking for Dynamic Token-Mixing Routing in Lightweight Image Super-Resolution

Image Super-Resolution (SR) aims to reconstruct high-resolution images from low-resolution counterparts, but the computational complexity of deep learning-based methods often hinders practical deployment. CAMixer is the pioneering work to integrate the advantages of existing lightweight SR methods and proposes a content-aware mixer to route token mixers of varied complexities according to the difficulty of content recovery. However, several limitations remain, such as poor adaptability, coarse-grained masking and spatial inflexibility, among others. We propose Pure-Pass (PP), a pixel-level masking mechanism that identifies pure pixels and exempts them from expensive computations. PP utilizes fixed color center points to classify pixels into distinct categories, enabling fine-grained, spatially flexible masking while maintaining adaptive flexibility. Integrated into the state-of-the-art ATD-light model, PP-ATD-light achieves superior SR performance with minimal overhead, outperforming CAMixer-ATD-light in reconstruction quality and parameter efficiency when saving a similar amount of computation.

cs.CV↗