WACV 2026 论文列表

Small Multimodal Models (SMMs) suffer under distribution shift after fine-tuning. Unsupervised Domain Adaptation (UDA) is a common remedy for this issue, but existing theory and methods are designed primarily for single- or dual-encoder architectures, overlooking the encoder-decoder structure of SMMs, whose fusion mechanism introduces additional shift. This work bridges this gap in two steps. First, we derive a dual-divergence risk bound that separates encoder divergence from fusion divergence and illustrate its tightness compared to the classical encoder-only bound with a negation-flip example. Second, motivated by this theory, we propose Dual-level Adversarial Alignment (DuAA), a two-stage alignment algorithm. DuAA inserts domain-discriminative adapters after the encoder and within the decoder to minimize both divergences. Furthermore, DuAA employs selective pseudo-labeling to refine target semantics. Our contribution targets domain shift in encoder-decoder SMMs and is agnostic to the fine-tuning mechanism, i.e., DuAA acts on internal representations, making it orthogonal to LoRA, or other fine-tuning variants. We adopt LoRA in experiments solely as a popular, parameter-efficient instantiation to keep the protocol fixed across settings. We compile twelve new cross-domain VQA tasks with distinct visual and textual shifts from existing datasets and observe that DuAA consistently outperforms standard fine-tuning across all tasks.

Text-to-image diffusion models may generate harmful or copyrighted content, motivating research on concept erasure. However, existing approaches primarily focus on erasing concepts from text prompts, overlooking other input modalities that are increasingly critical in real-world applications such as image editing and personalized generation. These modalities can become attack surfaces, where erased concepts re-emerge despite defenses. To bridge this gap, we introduce M-ErasureBench, a novel multimodal evaluation framework that systematically benchmarks concept erasure methods across three input modalities: text prompts, learned embeddings, and inverted latents. For the latter two, we evaluate both white-box and black-box access, yielding five evaluation scenarios. Our analysis shows that existing methods achieve strong erasure performance against text prompts but largely fail under learned embeddings and inverted latents, with Concept Reproduction Rate (CRR) exceeding 90% in the white-box setting. To address these vulnerabilities, we propose IRECE (Inference-time Robustness Enhancement for Concept Erasure), a plug-and-play module that localizes target concepts via cross-attention and perturbs the associated latents during denoising. Experiments demonstrate that IRECE consistently restores robustness, reducing CRR by up to 40% under the most challenging white-box latent inversion scenario, while preserving visual quality. To the best of our knowledge, M-ErasureBench provides the first comprehensive benchmark of concept erasure beyond text prompts. Together with IRECE, our benchmark offers practical safeguards for building more reliable protective generative models.

Intraoperative 2D/3D registration aligns preoperative 3D volumes with real-time 2D radiographs, enabling accurate localization of instruments and implants. A recent fully differentiable similarity learning framework approximates geodesic distances on SE(3), expanding the capture range of registration and mitigating the effects of substantial disturbances, but existing Euclidean approximations distort manifold structure and slow convergence. To address the above limitations, we explore similarity learning on non-Euclidean spherical feature spaces to improve the ability to capture and fit complex manifold features. We extract feature embeddings using a CNN-Transformer encoder, project them into spherical space, and approximate their geodesic distances with Riemannian geodesic distances in the bi-invariant SO(4) space. This enables the learning of a more expressive and geometrically consistent deep similarity metric, enhancing the network's ability to distinguish subtle pose differences. Fully differentiable Levenberg-Marquardt optimization is adopted to replace the existing gradient descent method to accelerate the convergence of the search during inference phase. Experiments on real and synthetic datasets show superior accuracy in both patient-specific and patient-agnostic scenarios.

General-purpose super-resolution models, particularly Vision Transformers, have achieved remarkable success but exhibit fundamental inefficiencies in common infrared imaging scenarios like surveillance and autonomous driving, which operate from fixed or nearly-static viewpoints. These models fail to exploit the strong, persistent spatial priors inherent in such scenes, leading to redundant learning and suboptimal performance. To address this, we propose the Regional Prior attention Transformer for infrared image Super-Resolution (RPT-SR), a novel architecture that explicitly encodes scene layout information into the attention mechanism. Our core contribution is a dual-token framework that fuses (1) learnable, regional prior tokens, which act as a persistent memory for the scene's global structure, with (2) local tokens that capture the frame-specific content of the current input. By utilizing these tokens into an attention, our model allows the priors to dynamically modulate the local reconstruction process. Extensive experiments validate our approach. While most prior works focus on a single infrared band, we demonstrate the broad applicability and versatility of RPT-SR by establishing new state-of-the-art performance across diverse datasets covering both Long-Wave (LWIR) and Short-Wave (SWIR) spectra. Our code will be publicly available.

We present PromptGAR, a novel framework for Group Activity Recognition (GAR) that offering both input flexibility and high recognition accuracy. The existing approaches suffer from limited real-world applicability due to their reliance on full prompt annotations, fixed number of frames and instances, and the lack of actor consistency. To bridge the gap, we proposed PromptGAR, which is the first GAR model to provide input flexibility across prompts, frames, and instances without the need for retraining. We leverage diverse visual prompts--like bounding boxes, skeletal keypoints, and instance identities--by unifying them as point prompts. A recognition decoder then cross-updates class and prompt tokens for enhanced performance. To ensure actor consistency for extended activity durations, we also introduce a relative instance attention mechanism that directly encodes instance identities. Comprehensive evaluations demonstrate that PromptGAR achieves competitive performances both on full prompts and partial prompt inputs, establishing its effectiveness on input flexibility and generalization ability for real-world applications.

In autonomous driving scenarios, accurate perception is becoming an even more critical task for safe navigation. While LiDAR provides precise spatial data, its inherent sparsity makes it difficult to detect small or distant objects. Existing methods try to address this by generating additional points within a Region of Interest (RoI), but relying on LiDAR alone often leads to false positives and a failure to recover meaningful structures. To address these limitations, we propose Image-Guided Semantic Pseudo-LiDAR Point Generation model, called ImagePG, a novel framework that leverages rich RGB image features to generate dense and semantically meaningful 3D points. Our framework includes an Image-Guided RoI Points Generation (IG-RPG) module, which creates pseudo-points guided by image features, and an Image-Aware Occupancy Prediction Network (I-OPN), which provides spatial priors to guide point placement. A multi-stage refinement (MR) module further enhances point quality and detection robustness. To the best of our knowledge, ImagePG is the first method to directly leverage image features for point generation. Extensive experiments on the KITTI and Waymo datasets demonstrate that ImagePG significantly improves the detection of small and distant objects like pedestrians and cyclists, reducing false positives by nearly 50%. On the KITTI benchmark, our framework improves mAP by +1.38%p (car), +7.91%p (pedestrian), and +5.21%p (cyclist) on the test set over the baseline, achieving state-of-the-art cyclist performance on the KITTI leaderboard. The code is available at: https://github.com/MS-LIMA/ImagePG

Video anomaly detection (VAD) is a crucial task for public safety and workforce reduction. Due to the rarity of abnormal events and the high cost of data collection, one-class classification (OCC) methods are extensively used. OCC methods are divided into object- and frame-centric approaches, each with its limitations. Object-centric methods fail to detect nonobject anomalies because they focus solely on objects, whereas frame-centric methods struggle to identify abnormalities due to a higher background rate than the foreground rate in video frames. To this end, we define three types of abnormal events, namely, human, appearance, and nonobject anomalies, and propose a unified VAD (UniVAD) model that effectively detects each defined anomaly type. UniVAD comprises three streams, namely, skeleton, local-visual, and global-visual, and each stream focuses on a specific type of anomaly. In addition, each stream uses an autoencoder; thus, we introduce the feature future past prediction task, which predicts past and future features based on present feature to suppress the strong generalization capacity of autoencoders. We validate the proposed model on three public benchmarks, ShanghaiTech, UBnormal, and NWPUCampus, and demonstrate that it achieves state-of-the-art performance by a significant margin.

Large-scale pre-trained video diffusion models have exhibited remarkable capabilities in diverse video generation. However, existing solutions face several challenges in generating long videos with rich human-scene interactions (HSI), including unrealistic dynamics and affordance, lack of subject identity preservation, and the need for expensive training. To this end, we propose GenHSI, a training-free method for controllable generation of long HSI videos with 3D awareness. Taking inspiration from movie animation, we subdivide the video synthesis into three stages: (1) script writing, (2) pre-visualization, and (3) animation. Given an image of a scene and a character with a user description, we use these three stages to generate long videos that preserve human identity and provide rich and plausible HSI. Script writing converts a complex text prompt involving a chain of HSI into simple atomic actions that are used in the pre-visualization stage to generate 3D keyframes. To synthesize plausible human interaction poses in 3D keyframes, we utilize pre-trained 2D inpainting diffusion models to generate plausible 2D human interactions based on view canonicalization, which eliminates the need for multi-view fitting in previous works. We then extend these interactions to 3D using robust iterative optimization, informed by contact cues and reasoning from VLMs. Prompted by these 3D keyframes, the pretrained video diffusion models can better generate consistent long videos with plausible dynamics and affordance in a 3D-aware manner. We are the first to synthesize a long video sequence with a chain of HSI actions without training based on the image references of the scene and character. Experiments demonstrate that our method can generate HSI videos that effectively preserve scene content and character identity with plausible human-scene interaction from a single image scene.

Despite the remarkable capabilities of Large Vision Language Models (LVLMs), they still lack detailed knowledge about specific entities. Retrieval-augmented Generation (RAG) is a widely adopted solution that enhances LVLMs by providing additional contexts from an external Knowledge Base. However, we observe that previous decoding methods for RAG are sub-optimal as they fail to sufficiently leverage multiple relevant contexts and suppress the negative effects of irrelevant contexts. To this end, we propose Relevance-aware Multi-context Contrastive Decoding (RMCD), a novel decoding method for RAG. RMCD outputs a final prediction by combining outputs predicted with each context, where each output is weighted based on its relevance to the question. By doing so, RMCD effectively aggregates useful information from multiple relevant contexts while also counteracting the negative effects of irrelevant ones. Experiments show that RMCD consistently outperforms other decoding methods across multiple LVLMs, achieving the best performance on three knowledge-intensive visual question-answering benchmarks. Also, RMCD can be simply applied by replacing the decoding method of LVLMs without additional training. Analyses also show that RMCD is robust to the retrieval results, consistently performing the best across the weakest to the strongest retrieval results.

Deep learning for medical imaging is hampered by task-specific models that lack generalizability and prognostic capabilities, while existing 'universal' approaches suffer from simplistic conditioning and poor medical semantic understanding. To address these limitations, we introduce DuPLUS, a deep learning framework for efficient multimodal medical image analysis. DuPLUS introduces a novel vision-language framework that leverages hierarchical semantic prompts for fine-grained control over the analysis task, a capability absent in prior universal models. To enable extensibility to other medical tasks, it includes a hierarchical, text-controlled architecture driven by a unique dual-prompt mechanism. For segmentation, DuPLUS is able to generalize across three imaging modalities, ten different anatomically various medical datasets, encompassing more than 30 organs and tumor types. It outperforms the state-of-the-art task-specific and universal models on 8 out of 10 datasets. We demonstrate extensibility of its text-controlled architecture by seamless integration of electronic health record (EHR) data for prognosis prediction, and on a head and neck cancer dataset, DuPLUS achieved a Concordance Index (CI) of 0.69. Parameter-efficient fine-tuning enables rapid adaptation to new tasks and modalities from varying centers, establishing DuPLUS as a versatile and clinically relevant solution for medical image analysis. The code for this work is made available at: https://anonymous.4open.science/r/DuPLUS-6C52

Recent advances in generative modeling enable image editing assistants that follow natural language instructions without additional user input. Their supervised training requires millions of triplets (original image, instruction, edited image), yet mining pixel-accurate examples is hard. Each edit must affect only prompt-specified regions, preserve stylistic coherence, respect physical plausibility, and retain visual appeal. The lack of robust automated edit-quality metrics hinders reliable automation at scale. We present an automated, modular pipeline that mines high-fidelity triplets across domains, resolutions, instruction complexities, and styles. Built on public generative models and running without human intervention, our system uses a task-tuned Gemini validator to score instruction adherence and aesthetics directly, removing any need for segmentation or grounding models. Inversion and compositional bootstrapping enlarge the mined set by approximately 2.6x, enabling large-scale high-fidelity training data. By automating the most repetitive annotation steps, the approach allows a new scale of training without human labeling effort. To democratize research in this resource-intensive area, we release NHR-Edit, an open dataset of 720k high-quality triplets, curated at industrial scale via millions of guided generations and validator passes, and we analyze the pipeline's stage-wise survival rates, providing a framework for estimating computational effort across different model stacks. In the largest cross-dataset evaluation, it surpasses all public alternatives. We also release Bagel-NHR-Edit, a fine-tuned Bagel model with state-of-the-art metrics. Project page, dataset, and model: https://riko0.github.io/No-Humans-Required/

Recent advances in foundational 3D reconstruction models, such as DUSt3R and MASt3R, have shown great potential in 2D and 3D correspondence in static scenes. In this paper, we propose to adapt them for the task of point tracking through 3D grounded correspondence. We first demonstrate that these models are competitive point trackers when focusing on static points, present in current point tracking benchmarks (+33.5% on EgoPoints vs. CoTracker2). We propose to combine the reconstruction loss with training for dynamic correspondence along with a visibility head, and fine-tuning MASt3R for point tracking using a relatively small amount of synthetic data. Importantly, we only train and evaluate on pairs of frames where one contains the query point, effectively removing any temporal context. Using a mix of dynamic and static point correspondences, we achieve competitive or superior point tracking results on four datasets (e.g. competitive on TAP-Vid-DAVIS 73.8 \delta_ avg / 85.8% occlusion acc. for PointSt3R compared to 75.7 / 88.3% for CoTracker2; and significantly outperform CoTracker3 on EgoPoints 61.3 vs 54.2 and RGB-S 87.0 vs 82.8). We also present results on 3D point tracking along with several ablations on training datasets and percentage of dynamic correspondences.

Recent learning-based visual localization methods use global descriptors to disambiguate visually similar places, but existing approaches often derive these descriptors from geometric cues alone (e.g., covisibility graphs), limiting their discriminative power and reducing robustness in the presence of noisy geometric constraints. We propose an aggregator module that learns global descriptors consistent with both geometrical structure and visual similarity, ensuring that images are close in descriptor space only when they are visually similar and spatially connected. This corrects erroneous associations caused by unreliable overlap scores. Using a batch-mining strategy based solely on the overlap scores and a modified contrastive loss, our method trains without manual place labels and generalizes across diverse environments. Experiments on challenging benchmarks show substantial localization gains in large-scale environments while preserving computational and memory efficiency. Code is github.com/sontung/robust_scr.

Person Re-Identification (ReID) task is important for designing intelligent surveillance systems. ReID can be highly challenging in low-light and low resolution scenarios. Existing ReID datasets predominantly feature cropped pedestrian images captured in well-lit environments, often lacking semantic richness, frame-level temporal continuity, and robustness to adverse conditions. To address these limitations, we introduce VAST-ReID, a new benchmark dataset specifically designed for the low-light person ReID task in real-world surveillance contexts. VAST-ReID consists of 1,211 surveillance videos collected at 21 different locations, capturing 169 distinct pedestrians of various age groups. The dataset emphasizes naturally low-light and visually degraded scenarios. Each identity is annotated with dense bounding boxes and enriched with auxiliary semantic labels, including pedestrian attributes and LLM-generated descriptions. While these annotations are not used during supervised training, they provide valuable semantic context for advancing research in language-guided retrieval and attribute-aware modeling. Additionally, we release identity-aligned image crops under the BoxTrack-ReID subset, which has over 14.3K frames sampled at 1fps from the raw videos, with standard training, gallery, and query splits compatible with the Market-1501 evaluation protocol, enabling straightforward benchmarking. The dataset has been benchmarked against SOTA methods, and experiments reveal that there is huge scope for improvement in ReID research.

Visual content memorability has intrigued the scientific community for decades, with applications ranging widely, from understanding nuanced aspects of human memory to enhancing content design. A significant challenge in progressing the field lies in the expensive process of collecting memorability annotations from humans. This limits the diversity and scalability of datasets for modeling visual content memorability. Most existing datasets are limited to collecting aggregate memorability scores for visual content, not capturing the nuanced memorability signals present in natural, open-ended recall descriptions. In this work, we introduce the first large-scale unsupervised dataset designed explicitly for modeling visual memorability signals, containing over 82,000 videos, accompanied by descriptive recall data. We leverage tip-of-the-tongue (ToT) retrieval queries from online platforms such as Reddit. We demonstrate that our unsupervised dataset provides rich signals for two memorability-related tasks: recall generation and ToT retrieval. Large vision-language models fine-tuned on our dataset outperform state-of-the-art models such as GPT-4o in generating open-ended memorability descriptions for visual content. We also employ a contrastive training strategy to create the first model capable of performing multimodal ToT retrieval. Our dataset and models present a novel direction, facilitating progress in visual content memorability research.

Scene text detection has progressed rapidly, largely driven by curated datasets and benchmarks. However, many of these have reached evaluation saturation and are heavily biased toward focused scenes, limiting their effectiveness in real-world environments where detection is hindered by environmental factors. To address this, we introduce MIST - a Multilingual Incidental Scene Text dataset featuring diverse text instances across 11 languages. MIST provides language, legibility, and fine-grained polygon-shaped annotations across 12K scene images and 600K word-level text instances. Images are captured along roads using a GoPro mounted on a moving car to capture real-world complexities, ensuring the scenes are incidental rather than deliberately framed. MIST establishes a new challenging benchmark to enable robust evaluation of scene text detection methods in real-world scenarios. The datasets and code will be available at https://saumya-svm.github.io/mist.

Understanding dynamic 4D environments--3D space evolving over time--is critical for robotic and interactive systems. These applications demand systems that can process streaming point cloud video in real-time, often under resource constraints, while also benefiting from past and present observations when available. However, current 4D backbone networks rely heavily on spatiotemporal convolutions and Transformers, which are often computationally intensive and poorly suited to real-time applications. We propose PointNet4D, a lightweight 4D backbone optimized for both online and offline settings. At its core is a Hybrid Mamba-Transformer temporal fusion block, which integrates the efficient state-space modeling of Mamba and the bidirectional modeling power of Transformers. This enables PointNet4D to handle variable-length online sequences efficiently across different deployment scenarios. To enhance temporal understanding, we introduce 4DMAP, a frame-wise masked auto-regressive pretraining strategy that captures motion cues across frames. Our extensive evaluations across 9 tasks on 7 datasets, demonstrating consistent improvements across diverse domains. We further demonstrate PointNet4D's utility by building two robotic application systems: 4D Diffusion Policy (DP4) and 4D Imitation Learning (4DIL), achieving substantial gains on the RoboTwin and HandoverSim benchmarks.

Real-time imaging sonar is crucial for underwater monitoring where optical sensing fails, but its use is limited by low uplink bandwidth and severe sonar-specific artifacts (speckle, motion blur, reverberation, acoustic shadows) affecting up to 98% of frames. We present SCOPE, a self-supervised framework that jointly performs compression and artifact correction without clean-noise pairs or synthetic assumptions. SCOPE combines (i) Adaptive Codebook Compression (ACC), which learns frequency-encoded latent representations tailored to imaging sonar, with (ii) Frequency-Aware Multiscale Segmentation (FAMS), which decomposes frames into low-frequency structure and sparse high-frequency dynamics while suppressing rapidly fluctuating artifacts. A hedging training strategy further guides frequency-aware learning using low-pass proxy pairs generated without labels. Evaluated on months of in-situ ARIS sonar data, SCOPE achieves a structural similarity index (SSIM) of 0.77, representing a 40% improvement over prior self-supervised denoising baselines, at bitrates down to \leq 0.0118 bpp. It reduces uplink bandwidth by more than 80% while improving downstream detection. The system runs in real time, with 3.1 ms encoding on an embedded GPU and 97 ms full multi-layer decoding on the server end. SCOPE has been deployed for months in three Pacific Northwest rivers to support real-time salmon enumeration and environmental monitoring in the wild. Results demonstrate that learning frequency-structured latents enables practical, low-bitrate sonar streaming with preserved signal details under real-world deployment conditions.

Training a deep neural network with a small amount of labeled data is challenging. The challenge is even more severe for medical images because of the many possible variations in the images. We propose a novel framework for few-shot chest x-ray (CXR) diagnosis. For classification problems, training with limited data may be facilitated if class-specific features can be extracted and utilized. Semantic information about the abnormalities may also be helpful in this context. To that end, we design an autoencoder-based approach that extracts visual features and decomposes them into class-agnostic and class-specific features utilizing the semantic information of the abnormalities. The decomposition helps in efficient classification using the class-specific features. Additionally, we perform test-time adaptation to deal with possible variations in the test data compared to the training data. From this perspective, our method is one of the first of its kind. Extensive evaluations on publicly available chest x-ray datasets under few-shot settings show the effectiveness of our method. Results on the publicly available chest x-ray datasets show a 3-5% improvement in AUROC scores.