WACV 2026 Papers

Vision-language models (VLMs) often inherit the biases and unsafe associations present within their large-scale training dataset. While recent approaches mitigate unsafe behaviors, their evaluation focuses on how safe the model is on unsafe inputs, ignoring potential shortcomings on safe ones. In this paper, we first revise safety evaluation by introducing SafeGround, a new set of metrics that evaluate safety at different levels of granularity. With this metric, we uncover a surprising issue of training-based methods: they make the model less safe on safe inputs. From this finding, we take a different direction and explore whether it is possible to make a model safer without training, introducing Unsafe Weights Manipulation (UWM). UWM uses a calibration set of safe and unsafe instances to compare activations between safe and unsafe content, identifying the most important parameters for processing the latter. Their values are then manipulated via negation. Experiments show that UWM achieves the best tradeoff between safety and knowledge preservation, consistently improving VLMs on unsafe queries while outperforming even training-based state-of-the-art methods on safe ones.

Accurately predicting directional radio maps is essential for wireless applications, yet prior approaches primarily focus on omnidirectional signals and typically treat transmitter localization and signal map reconstruction as separate tasks. In omnidirectional settings, predicting the maximum signal location often coincides with the transmitter position, which limits the need for explicit joint modeling. However, in directional propagation where angular effects, reflections, and building occlusions play critical roles, this assumption no longer holds. To address this gap, we propose SymNet, a unified framework that jointly predicts directional radio maps and transmitter locations from sparse signal measurements. SymNet incorporates a prediction head for transmitter localization alongside radio map reconstruction, enabling simultaneous learning of both tasks. This joint formulation leverages their complementary information and leads to consistent improvements over treating them separately. Experiments on challenging directional scenarios demonstrate that SymNet outperforms state-of-the-art baselines, achieving superior accuracy in both radio map reconstruction and transmitter localization.

Realistic scene simulation is a promising way to improve autonomous driving. While existing diffusion-based 2D augmentation and 3D asset libraries show potential for synthesizing diverse driving scenarios, they often struggle with multi-view photorealistic rendering and consistency. These issues are particularly challenging for vehicle-to-everything (V2X) collaborative perception, since its effectiveness relies on precise geometric alignment and visual coherence across multiple viewpoints. To address these challenges, we propose V2XScene, a 3D driving scene editing framework. This framework enhances V2X collaborative perception by enabling high-quality 3D vehicle asset generation and consistent multi-view insertion. V2XScene consists of three components: a visual question answering (VQA) guided generation module for query-driven 3D vehicle asset synthesis; a 3D object mapping module for vehicle placement optimization and occlusion reasoning; and a realistic insertion module for lighting estimation and virtual vehicle insertion. Extensive experiments demonstrate that V2XScene can generate multi-view consistent and realistic driving scenes, which significantly improves V2X perception accuracy.

We present NRGMark, a region-based image watermarking framework to embed energy use and provenance metadata into images. Targeting composite graphic designs such as posters, NRGMark enables imperceptible watermarking of distinct visual elements each carrying independent metadata on their environmental impact, such as the energy consumption associated with generative AI (GenAI) use. NRGMark extends image watermark encoder-decoder models by incorporating an object localization network to detect and decode multiple watermarked regions within a document, even under image transformations and physical print-scan degradation. NRGMark interoperates with several watermarking techniques and the emerging C2PA open standard for media provenance to encode environmental impact metadata. We demonstrate NRGMark on both synthetic and real-world design layouts, illustrating its potential to support energy transparency in the age of GenAI.

Vision-language models (VLMs) have demonstrated substantial promise in medical image segmentation by utilizing radiology reports as prompts to segment regions of interest. However, VLM deployment in clinical settings is challenged by two intertwined issues: i) demographic bias, where performance varies across demographic groups, and ii) prompt sensitivity, where semantically similar prompts yield inconsistent outputs. These challenges are interconnected; demographic underrepresentation can worsen a model's sensitivity to prompts, and prompt instability can more heavily affect certain demographic groups. In this study, we present FairVLM, a unified framework that addresses both demographic disparity and prompt sensitivity in VLMs. FairVLM integrates three key components: (1) Semantic-Retaining Counterfactual Prompting (SRCP), which generates clinically consistent and diverse prompt variations via large language models; (2) Demographic-Aware Feature Normalization (DAFN), a lightweight module that mitigates latent representation bias across demographic groups; and (3) a Fairness-Calibrated Loss (FCL) that explicitly penalizes performance disparities while encouraging prompt consistency. Extensive evaluations on the Harvard-FairSeg dataset show that FairVLM significantly improves equity-scaled segmentation. It also reduces demographic disparity (DI) by over 65% and relative performance gap (RPG) by over 60%, while maintaining or boosting overall accuracy. FairVLM is robust to prompt changes, with less than 0.5% performance drop across varied prompts, and also generalizes well on unseen datasets. These findings present FairVLM as a new state-of-the-art, robust, and adaptable framework for a fair, prompt-invariant vision-language model. Code and data are available at https://github.com/Rahman-Motiur/FairVLM.

Existing vision-language model (VLM)-based methods for out-of-distribution (OOD) detection typically rely on similarity scores between input images and in-distribution (ID) text prototypes. However, the modality gap between image and text often results in high false positive rates, as OOD samples can exhibit high similarity to ID text prototypes. To mitigate the impact of this modality gap, we propose incorporating ID image prototypes along with ID text prototypes. We present theoretical and empirical evidence indicating that this approach enhances VLM-based OOD detection performance without any additional training. To further reduce the gap between image and text, we introduce a novel few-shot tuning framework, suPreMe, comprising biased prompt generation (BPG) and image-text consistency (ITC) modules. BPG enhances image-text fusion and improves generalization (prevents overfitting on training data) by conditioning ID text prototypes on the Gaussian-based estimated image domain bias; ITC reduces the modality gap by minimizing intra- and inter-modal distances. Moreover, inspired by our theoretical and empirical findings, we introduce a novel OOD score S_ GMP , leveraging uni- and cross-modal similarities. Finally, we present extensive experiments to demonstrate that suPreMe consistently outperforms existing VLM-based OOD detection methods.

3D scene understanding is a critical yet challenging task in autonomous driving due to the irregularity and sparsity of LiDAR data, as well as the computational demands of processing large-scale point clouds. Recent methods leverage range-view representations to enhance efficiency, but they often adopt higher azimuth resolutions to mitigate information loss during spherical projection, where only the closest point is retained for each 2D grid. However, processing wide panoramic range-view images remains inefficient and may introduce additional distortions. Our empirical analysis shows that training with multiple range images, obtained from splitting the full point cloud, improves both segmentation accuracy and computational efficiency. However, this approach also poses new challenges of exacerbated class imbalance and increase in projection artifacts. To address these, we introduce FLARES, a novel training paradigm that incorporates two tailored data augmentation techniques and a specialized post-processing method designed for multi-range settings. Extensive experiments demonstrate that FLARES is highly generalizable across different architectures, yielding 2.1%-7.9% mIoU improvements on SemanticKITTI and 1.8%-3.9% mIoU on nuScenes, while delivering over 40% speed-up in inference.

Monocular 3D foundation models offer an extensible solution for perception tasks, making them attractive for broader 3D vision applications.In this paper, we propose MoRe, a training-free Monocular Geometry Refinement method designed to improve cross-view consistency and achieve scale alignment.To induce inter-frame relationships, our method employs feature matching between frames to establish correspondences.Rather than applying simple least squares optimization on these matched points, we formulate a graph-based optimization framework that performs local planar approximation using the estimated 3D points and surface normals estimated by monocular foundation models.This formulation addresses the scale ambiguity inherent in monocular geometric priors while preserving the underlying 3D structure.We further demonstrate that MoRe not only enhances 3D reconstruction but also improves novel view synthesis, particularly in sparse-view rendering scenarios.

Underwater object detection is critical for monitoring marine ecosystems but poses unique challenges, including degraded image quality, imbalanced class distribution, and distinct visual characteristics. Not every species is detected equally well, yet underlying causes remain unclear. We investigate two questions: 1) What factors beyond data quantity drive class-specific performance disparities? 2) How can we systematically improve detection of under-performing marine species? Using manipulated DUO and RUOD datasets, we decouple localization from classification and investigate the scallop class as a representative case. Localization analysis using YOLO11 and TIDE shows that foreground-background discrimination is the most problematic stage regardless of data quantity. Classification experiments reveal persistent precision gaps even under balanced sampling, indicating intrinsic feature-based challenges beyond data scarcity. We recommend imbalanced distributions when prioritizing precision, and balanced distributions when prioritizing recall. Improving under-performing classes should focus on algorithmic advances, especially within localization modules. We publicly release our code and datasets.

Recent advances in action anticipation rely heavily on Transformer architectures to learn discriminative representations of the past observation, incurring high computational and memory overhead that limits their applicability to long videos. While temporal processors with linear complexity like RNNs and state-space models offer efficient alternatives, their sequential nature risks overlooking subtle cues in observed frames that could enhance future anticipation. We address this limitation with Cross Linear Attentive Memory (CLAM), a memory module that selectively retrieves complementary context cues from frame features. By reformulating linear attention to replace traditional cross-attention, CLAM achieves linear computation complexity and constant memory usage relative to input length. Finally, by fusing the outputs of the temporal processor and CLAM, a non-autoregressive Transformer decoder generates future actions in one shot with high accuracy. Experiments on egocentric (EpicKitchens100 and Ego4D) and third-person (Thumos14) benchmarks demonstrate our model's superior anticipation accuracy and scalability, processing longer sequences with significantly less latency growth than alternatives. Our approach also achieves promising results in online action detection.

Deep learning models achieve high predictive performance but lack intrinsic interpretability, hindering our understanding of the learned prediction behavior. Existing local explainability methods focus on associations, neglecting the causal drivers of model predictions. Other approaches adopt a causal perspective but primarily provide global, model-level explanations. However, for specific inputs, it's unclear whether globally identified factors apply locally. To address this limitation, we introduce a novel framework for local interventional explanations by leveraging recent advances in image-to-image editing models. Our approach performs gradual interventions on semantic properties to quantify the corresponding impact on a model's predictions using a novel score, the expected property gradient magnitude. We demonstrate the effectiveness of our approach through an extensive empirical evaluation on a wide range of architectures and tasks. First, we validate it in a synthetic scenario and demonstrate its ability to locally identify biases. Afterward, we apply our approach to investigate medical skin lesion classifiers, analyze network training dynamics, and study a pre-trained CLIP model with real-life interventional data. Our results highlight the potential of interventional explanations on the property level to reveal new insights into the behavior of deep models.

Echocardiography, a widely used imaging modality, offers real-time assessments of cardiac morphology and function, with a particular emphasis on left ventricular dynamics. Despite its clinical importance, existing automated methods for echocardiographic analysis struggle to ensure temporal consistency in left ventricular key-point trajectories, largely due to their reliance on static frame annotations. To overcome these challenges, we propose a semi-supervised trajectory refinement framework that employs inter-frame correlations to enhance key-point estimation across echocardiography videos. A semi-supervised trajectory learning scheme is presented to improve the efficacy of key-point trajectory analysis using unannotated echocardiography videos. The experiments present considerable improvements in both spatial accuracy and temporal stability of the left ventricle key-point trajectories, outperforming state-of-the-art baselines and demonstrating the clinical applicability for robust echocardiography analysis.

Small Object Detection (SOD) is essential for real-world applications, including satellite image analysis and drone-based surveillance, where target objects typically exhibit limited spatial extent, high density, and visual similarity to complex backgrounds. These factors substantially hinder conventional object detection methods, resulting in weak feature representations and poor detection accuracy. To overcome these challenges, we introduce Segmentation-guided and Prior-Aided Routing for Small Object Detection (SPAR-Det), a unified framework integrating segmentation-guided attention, geometric prior supervision, and adaptive feature routing. At its core, SPAR-Det employs a Cross-Attention Heterogeneous Feature Fusion (CAHF) module that leverages pretrained segmentation backbones to enhance foreground object features while effectively suppressing background. Additionally, we propose a Geometric Prior Supervision Loss that combines Gaussian bounding box maps with segmentation feature maps, providing crucial geometric context and semantic cues to address the limited self-representation capability of small objects. Furthermore, our framework includes a Mixture-of-Experts (MoE) detection head, dynamically allocating specialized classifiers according to varying scene characteristics, thereby significantly improving generalization across diverse environments. Extensive evaluations conducted on three benchmark datasets, AI-TOD, Visdrone and LEVIR-Ship, demonstrate that SPAR-Det achieves state-of-the-art performance, verifying its robustness and applicability for challenging small object detection scenarios. Code is available at https://github.com/kwonseungchan/SPAR-Det.git

Scene Graph Generation (SGG) suffers from a long-tailed distribution, where a few predicate classes dominate while many others are underrepresented, leading to biased models that underperform on rare relations. Unbiased-SGG methods address this by implementing debiasing strategies, but often at the cost of spatial understanding--resulting in over-reliance on semantic priors. We introduce Salience-SGG, a novel framework featuring an Iterative Salience Decoder (ISD) that emphasizes triplets with salient spatial structures. To support this, we propose semantic-agnostic salience labels guiding ISD. Evaluations on Visual Genome, Open Images V6, and GQA-200 show that Salience-SGG achieves state-of-the-art performance and improves existing Unbiased-SGG methods in their spatial understanding as demonstrated by the Pairwise Localization Average Precision

Deep neural networks are powerful, yet their high complexity greatly limits their potential to be deployed on billions of resource-constrained edge devices. Pruning is a crucial network compression technique, yet most existing methods focus on classification models, with limited attention to detection. Even among those addressing detection, there is a lack of utilization of essential localization information. Also, many pruning methods passively rely on pre-trained models, in which useful and useless components are intertwined, making it difficult to remove the latter without harming the former at the neuron/filter level. To address the above issues, in this paper, we propose a proactive detection-discriminants-based network compression approach for deep visual detectors, which alternates between two steps: (1) maximizing and compressing detection-related discriminants and aligning them with a subset of neurons/filters immediately before the detection head, and (2) tracing the detection-related discriminating power across the layers and discarding features of lower importance. Object location information is exploited in both steps. Extensive experiments, employing four advanced detection models and four state-of-the-art competing methods on the KITTI and COCO datasets, highlight the superiority of our approach. Remarkably, our compressed models can even beat the original base models with a substantial reduction in complexity.

Episodic memory retrieval enables wearable cameras to recall objects or events previously observed in video.However, existing formulations assume an "offline" setting with full video access at query time, limiting their applicability in real-world scenarios with power and storage-constrained wearable devices.Towards more application-ready episodic memory systems, we introduce Online Visual Query 2D (OVQ2D), a task where models process video streams online, observing each frame only once, and retrieve object localizations using a compact memory instead of full video history.We address OVQ2D with ESOM (Egocentric Streaming Object Memory), a novel framework integrating an object discovery module, an object tracking module, and a memory module that find, track, and store spatio-temporal object information for efficient querying.Experiments on Ego4D demonstrate ESOM's superiority over other online approaches, though OVQ2D remains challenging, with top performance at only 4% success. ESOM's accuracy increases markedly with perfect object tracking (31.91%), discovery (40.55%), or both (81.92%), underscoring the need of applied research on these components.

In this work we present a novel approach for unsupervised multi-graph matching, which applies to problems for which a Gaussian distribution of keypoint features can be assumed. We leverage cycle consistency as loss for self-supervised learning, and determine Gaussian parameters through Bayesian Optimization, yielding a highly efficient approach that scales to large datasets. Our fully unsupervised approach enables us to reach the accuracy of state-of-the-art supervised methodology for the biomedical use case of semantic cell annotation in 3D microscopy images of the worm C. elegans. To this end, our approach yields the first unsupervised atlas of C. elegans, i.e. a model of the joint distribution of all of its cell nuclei, without the need for any ground truth cell annotation. This advancement enables highly efficient semantic annotation of cells in large microscopy datasets, overcoming a current key bottleneck. Beyond C. elegans, our approach offers fully unsupervised construction of cell-level atlases for any model organism with a stereotyped body plan down to the level of unique semantic cell labels, and thus bears the potential to catalyze respective biomedical studies in a range of further species.

Text-to-image (T2I) diffusion models have achieved widespread success due to their ability to generate high-resolution, photorealistic images. These models are trained on large-scale datasets, like LAION-5B, often scraped from the internet. However, since this data contains numerous biases, the models inherently learn and reproduce them, resulting in stereotypical outputs. We introduce SelfDebias, a fully unsupervised test-time debiasing method applicable to any diffusion model that uses a UNet as its noise predictor. SelfDebias identifies semantic clusters in an image encoder's embedding space and uses these clusters to guide the diffusion process during inference, minimizing the KL divergence between the output distribution and the uniform distribution. Unlike supervised approaches, SelfDebias does not require human-annotated datasets or external classifiers trained for each generated concept. Instead, it is designed to automatically identify semantic modes. Extensive experiments show that SelfDebias generalizes across prompts and diffusion model architectures, including both conditional and unconditional models. It not only effectively debiases images along key demographic dimensions while maintaining the visual fidelity of the generated images, but also more abstract concepts for which identifying biases is also challenging.

Human pose estimators are typically retrained from scratch or naively fine-tuned whenever keypoint sets, sensing modalities, or deployment domains change--an inefficient, compute-intensive practice that rarely matches field constraints. We present PoseAdapt, an open-source framework and benchmark suite for continual pose model adaptation. PoseAdapt defines domain-incremental and class-incremental tracks that simulate realistic changes in density, lighting, and sensing modality, as well as skeleton growth. The toolkit supports two workflows: (i) Strategy Benchmarking, which lets researchers implement continual learning (CL) methods as plugins and evaluate them under standardized protocols; and (ii) Model Adaptation, which allows practitioners to adapt strong pretrained models to new tasks with minimal supervision. We evaluate representative regularization-based methods in single-step and sequential settings. Benchmarks enforce a fixed lightweight backbone, no access to past data, and tight per-step budgets. This isolates adaptation strategy effects, highlighting the difficulty of maintaining accuracy under strict resource limits. PoseAdapt connects modern CL techniques with practical pose estimation needs, enabling adaptable models that improve over time without repeated full retraining.