The Segment Anything Model 2 (SAM 2) has demonstrated exceptional performance in object segmentation tasks but encounters challenges in visual object tracking, particularly in handling crowded scenes with fast-moving or self-occluding objects. Additionally, its fixed-window memory mechanism indiscriminately retains past frames, leading to error accumulation. This issue results in incorrect memory retention during occlusions, causing the model to condition future predictions on unreliable features and leading to identity switches or drift in crowded scenes. This paper introduces SAMURAI, an enhanced adaptation of SAM 2 that integrates temporal motion cues with a novel motion-aware memory selection strategy. SAMURAI effectively predicts object motion and refines mask selection, achieving robust and precise tracking without requiring retraining or fine-tuning. It demonstrates strong training-free performance across multiple VOT benchmark datasets, underscoring its generalization capability. SAMURAI achieves state-of-the-art performance on LaSOText, GOT-10k, and TrackingNet, while also delivering competitive results on LaSOT, VOT2020-ST, VOT2022-ST, and VOS benchmarks such as SA-V. These results highlight SAMURAI's robustness in complex tracking scenarios and its potential for real-world applications in dynamic environments with an optimized memory selection mechanism. Code and results are available at https://github.com/yangchris11/samurai.

This paper presents the outcomes of the PETS2025 challenge, held in conjunction with AVSS 2025 and sponsored by the EU-funded EURMARS project. The challenge introduces a novel maritime surveillance dataset comprising image sequences captured by diverse multi-altitude, multimodal sensors, reflecting the real-world multi-authority environment. The key tasks include: (1) object detection using various sensors across different platforms (ground-based and low-altitude aerial) and spectral ranges (visible, thermal, ultraviolet (UV), and short-wave infrared (SWIR)); (2) long-term tracking of targets in maritime environments spanning both sea and land; and (3) approximating target geolocations by using sensor imagery and telemetry data. Performance evaluations of results submitted by 12 international participants are discussed. The results show the effectiveness of these submissions and highlight ongoing challenges posed by heterogeneous sensors and complex environments. These challenges emphasise the need to further improve detection, tracking, and geolocation approximation for maritime and coastal surveillance.

Occlusion remains a major challenge in online Multi-Object Tracking (MOT), where existing multi-stage association methods often rely on detection confidence scores despite their weak correlation with occlusion, leading to frequent errors. We propose DARUMA, a depth-aware MOT framework that prioritizes non-occluded objects using occlusion-aware association by re-prioritizing the matching order and refines association with a depth-weighted cost metric for improved robustness in occluded and depth-varying environments. Additionally, we introduce Generic Observation-Centric Momentum (GOCM), which integrates depth-aware velocity estimation and confidence-weighted historical observations to enhance motion modeling. Our method can integrates into existing MOT frameworks, improving association robustness without additional supervision.Extensive evaluations on DanceTrack demonstrate that DARUMA achieves state-of-the-art performance, particularly in complex, occlusion-heavy scenarios.

Recent advancements in 3D Large Multi-modal Models (3D-LMMs) have driven significant progress in 3D question answering. However, recent multi-frame Vision-Language Models (VLMs) demonstrate superior performance compared to 3D-LMMs on 3D question answering tasks, largely due to the greater scale and diversity of available 2D image data in contrast to the more limited 3D data. Multi-frame VLMs, although achieving superior performance, suffer from the difficulty of retaining all the detailed visual information in the 3D scene while limiting the number of visual tokens. Common methods such as token pooling, reduce visual token usage but often lead to information loss, impairing the model’s ability to preserve visual details essential for 3D question answering tasks. To address this, we propose voxel-based Dynamic Token Compression (DTC), which combines 3D spatial priors and visual semantics to achieve over 90% reduction in visual tokens usage for current multi-frame VLMs. Our method maintains performance comparable to state-of-the-art models on 3D question answering benchmarks including OpenEQA and ScanQA, demonstrating its effectiveness.

In the field of multi-object tracking (MOT), traditional methods often rely on the Kalman filter for motion prediction, leveraging its strengths in linear motion scenarios. However, the inherent limitations of these methods become evident when confronted with complex, nonlinear motions and occlusions prevalent in dynamic environments like sports and dance. This paper explores the possibilities of replacing the Kalman filter with a learning-based motion model that effectively enhances tracking accuracy and adaptability beyond the constraints of Kalman filter-based tracker. In this paper, our proposed method MambaMOT and MambaMOT+, demonstrate advanced performance on challenging MOT datasets such as DanceTrack and SportsMOT, showcasing their ability to handle intricate, nonlinear motion patterns and frequent occlusions more effectively than traditional methods.

Unlike closed-vocabulary 3D instance segmentation that is often trained end-to-end, open-vocabulary 3D instance segmentation (OV-3DIS) often leverages vision-language models (VLMs) to generate 3D instance proposals and classify them. While various concepts have been proposed from existing research, we observe that these individual concepts are not mutually exclusive but complementary. In this paper, we propose a new state-of-the-art solution for OV-3DIS by carefully designing a recipe to combine the concepts together and refining them to address key challenges. Our solution follows the two-stage scheme: 3D proposal generation and instance classification. We employ robust 3D tracking-based proposal aggregation to generate 3D proposals and remove overlapped or partial proposals by iterative merging/removal. For the classification stage, we replace the standard CLIP model with Alpha-CLIP, which incorporates object masks as an alpha channel to reduce background noise and obtain object-centric representation. Additionally, we introduce the standardized maximum similarity (SMS) score to normalize text-to-proposal similarity, effectively filtering out false positives and boosting precision. Our framework achieves state-of-the-art performance on ScanNet200 and S3DIS across all AP and AR metrics, even surpassing an end-to-end closed-vocabulary method.

Token merging has emerged as an effective strategy to accelerate Vision Transformers (ViT) by reducing computational costs. However, existing methods primarily rely on the visual token’s feature similarity for token merging, overlooking the potential of integrating spatial information, which can serve as a reliable criterion for token merging in the early layers of ViT, where the visual tokens only possess weak visual information. In this paper, we propose ToSA, a novel token merging method that combines both semantic and spatial awareness to guide the token merging process. ToSA leverages the depth image as input to generate pseudo spatial tokens, which serve as auxiliary spatial information for the visual token merging process. With the introduced spatial awareness, ToSA achieves a more informed merging strategy that better preserves critical scene structure. Experimental results demonstrate that ToSA outperforms previous token merging methods across multiple benchmarks on visual and embodied question answering while largely reducing the runtime of the ViT, making it an efficient solution for ViT acceleration. The code will be available at: https://github.com/hsiangwei0903/ToSA.

In the dynamic and evolving field of computer vision, action recognition has become a key focus, especially with the advent of sophisticated methodologies like Convolutional Neural Networks (CNNs), Convolutional 3D, Transformer and spatial-temporal feature fusion. These technologies have shown promising results on well-established benchmarks but face unique challenges in real-world applications, particularly in sports analysis, where the precise decomposition of activities and the distinction of subtly different actions are crucial. Existing datasets like UCF101, HMDB51, and Kinetics have offered a diverse range of video data for various scenarios. However, there’s an increasing need for fine-grained video datasets that capture detailed categorizations and nuances within broader action categories. In this paper, we introduce the VideoBadminton dataset, which is derived from high-quality badminton footage. Through an exhaustive evaluation of leading methodologies on this dataset, this study aims to advance the field of action recognition, particularly in badminton sports. The introduction of VideoBadminton could not only serve for badminton action recognition but also provide a dataset for recognizing fine-grained actions. The insights gained from these evaluations are expected to catalyze further research in action comprehension, especially within sports contexts.

Dense object counting or crowd counting has come a long way thanks to the recent development in the vision community. However, indiscernible object counting, which aims to count the number of targets that are blended with respect to their surroundings, has been a challenge. Image-based object counting datasets have been the mainstream of the current publicly available datasets. Therefore, we propose a large-scale dataset called YoutubeFish-35, which contains a total of 35 sequences of high-definition videos with high frame-per-second and more than 159,000 annotated center points across a selected variety of scenes. For bench-marking purposes, we select three mainstream methods for dense object counting and carefully evaluate them on the newly collected dataset. We propose TransVidCount, a new strong baseline that combines density and regression branches along the temporal domain in a unified framework and can effectively tackle indiscernible object counting with state-of-the-art performance on YoutubeFish-35 dataset.

In the domain of autonomous driving, the integration of multi-modal perception techniques based on data from diverse sensors has demonstrated substantial progress. Effectively surpassing the capabilities of state-of-the-art single-modality detectors through sensor fusion remains an active challenge. This work leverages the respective advantages of cameras in perspective view and radars in Bird’s Eye View (BEV) to greatly enhance overall detection and tracking performance. Our approach, Camera-Radar Associated Fusion Tracking Booster (CRAFTBooster) represents a pioneering effort to enhance radar-camera fusion in the tracking stage, contributing to improved 3D MOT accuracy. The superior experimental results on K-Radaar dataset, which exhibit 5-6% on IDF1 tracking performance gain, validate the potential of effective sensor fusion in advancing autonomous driving.