This paper presents a cascaded Low-Rank Adaptation (LoRA)-based multimodal fusion framework for action and activity recognition in healthcare-oriented training environments. The proposed architecture combines parameter-efficient modality-specific adaptation with sequential fusion, enabling modalities to be integrated in stages without retraining previously learned components. Rather than assuming a fixed fusion structure, the framework first integrates more closely related modalities and then incorporates additional heterogeneous modalities, supporting scalable adaptation across datasets with different modality this http URL evaluate the framework on two healthcare-oriented training environment datasets: NurViD and the Nurse Training dataset. Across these datasets, preliminary results suggest that the proposed cascaded fusion strategy improves over individual modality models and provides competitive performance relative to previously reported dataset-specific baselines. Overall, these findings indicate that cascaded LoRA-based fusion is a promising parameter-efficient approach for integrating heterogeneous modalities in medical training action and activity recognition tasks. github: this https URL.
https://arxiv.org/abs/2607.11839
Precise Event Spotting (PES) requires distinguishing visually similar yet semantically distinct adjacent frames, making it fundamentally different from image classification and coarse action recognition. Although self-distillation methods such as DINO have shown strong representation learning ability in images, we find that directly applying them to PES is ineffective: without supervised guidance, subtle but crucial motion cues are often suppressed as noise, leading to representations that are insensitive to precise event boundaries. To address this, we propose Temporal Feature Distillation, a semi-supervised objective that aligns temporally informative backbone features, rather than projection-head outputs, to preserve motion-sensitive and boundary-aware cues for frame-level localization. A supervised warm-up with a ramp-up schedule further stabilizes training by ensuring that meaningful event cues are learned before unlabeled distillation begins. We also introduce Transformer Gate Shift, a multi-scale gated shifting module that injects motion-aware temporal information into Vision Transformers. Experiments on four fine-grained sports benchmarks show consistent improvements over fully supervised and semi-supervised baselines. Under 10\% supervision on FSPerf, our method improves mAP by 4.54 points over the strongest competing approach, and with only 80\% labeled data, it matches or surpasses the fully supervised 100\% baseline on two of the four datasets.
https://arxiv.org/abs/2607.10998
Assembly action understanding is a key enabler for effective human-robot collaborative assembly, yet it remains challenging due to subtle motions and fine-grained hand-object interactions. We adapt vision-language models (VLMs) to this challenging domain with Compositional Context Fine-Tuning (CCFT), a method that decomposes assembly actions into semantic elements (Verb, Object, Tool) and fine-tunes VLMs to recognize each action element using templated question-answering pairs. This approach ensures near-deterministic outputs. To enable efficient and effective multi-task learning under limited data, a Layer-Partitioned Alternating Training (LP-AT) method is presented, which assigns distinct model layers to recognize specific action elements through element-specific low-rank adapters. LP-AT alternates weight updates across element-specific adapters, reducing cross-task interference while enabling per-adapter hyperparameter optimization. Furthermore, we create HA-ViD-VQA and IKEA-ASM-VQA datasets from existing assembly video datasets. Extensive experiments on these datasets demonstrate that our method consistently outperforms strong action recognition baselines while providing interpretable element-level predictions that can support diverse downstream applications.
https://arxiv.org/abs/2607.10797
Micro-Actions (MAs) are subtle and spontaneous human behaviors that provide important non-verbal cues in social interaction and affective communication. However, their short duration, weak motion patterns, and fine-grained semantic differences make them difficult to annotate, model, and evaluate in a standardized manner. To promote academic research on micro-action analysis, we proposed and have annually organized the Micro-Action Analysis Grand Challenge (MAC) as a public benchmark platform for this emerging field. The first two editions of MAC established standardized evaluation settings for micro-action recognition and detection, providing publicly accessible datasets and protocols. Building upon these editions, this paper presents the 3rd MAC, held in conjunction with ACM Multimedia 2026. Under the theme of moving from recognition to fine-grained micro-action understanding, this edition further expands the scope of the challenge beyond conventional recognition and detection. In particular, we introduce a new task named fine-grained micro-action understanding, evaluated with the assistance of multimodal large language models, aiming to assess models' ability to capture fine-grained semantic cues and interpret subtle human micro-actions at a deeper level. We summarize the datasets, task settings, evaluation protocols, competition results, and representative solutions from top-performing teams. Finally, we discuss future directions for micro-action analysis and its broader role in human-centric video understanding.
https://arxiv.org/abs/2607.16284
Hand-object interaction (HOI) recognition requires capturing both hand manipulations and object transformations. However, existing video-language models often fall into shortcuts by relying on spurious correlations among hands, objects, or environmental context, rather than reasoning from the appearance and dynamics of hands and objects themselves. To address this limitation, we propose a new learning paradigm that combines (i) hand-object masked training, which enables robust reasoning from partial hand or object observations, and (ii) an HOI-dynamics-aware decoder that explicitly learns hand- and object-centric embeddings through auxiliary predictions of their locations and semantics, enhancing sensitivity to both cues. To systematically evaluate such cue-specific reasoning, we introduce Cue-Isolated HOI (CI-HOI), a new evaluation that assesses models' ability to predict actions from hand- and object-related cues independently. To enable CI-HOI, we curate the DEHOI testbed, which separates hand- and object-related observations for disentangled HOI evaluation through inpainting. Using DEHOI, we demonstrate both quantitatively and qualitatively that our training strategy exploits hand- and object-centric information more effectively than existing models. Our approach improves over existing models on DEHOI, standard action recognition, object state recognition, and even robot manipulation action recognition, leading to more robust HOI understanding.
https://arxiv.org/abs/2607.08514
Existing video benchmarks evaluate action recognition on consumer videos, egocentric recordings, or simulated industrial environments. They do not test vision-language models under the visual and procedural conditions of real industrial CCTV, where workers appear as distant figures amid dust, steam, low light, glare, occlusion, and overlapping activities. We introduce STEELBENCH, a diagnostic benchmark for industrial surveillance that jointly evaluates per-worker activity recognition, safety-rule reasoning, and annotation provenance. SteelBench contains 1,345 densely annotated clips, curated from 149 hours of operational plant footage and 10,024 candidate clips using temporal deduplication, class balancing, and visibility-aware stratified sampling. Each clip includes dense per-worker action labels, PPE attributes, spatial context, and safety-rule annotations. Because model-assisted annotation can shape the labels later used for model evaluation, SteelBench includes a provenance-aware audit protocol. The protocol measures label influence, evaluates sensitivity to ground-truth provenance, and reports a human reference from expert-reviewed labels. Applying this audit, we find that unaudited VLM-sourced ground truth can inflate same-family model accuracy by up to 17 percentage points. Across nine VLMs from four architectural families, the best model reaches only 42.6% action accuracy, compared with an 84.6% human benchmark. Performance also fragments across recognition, robustness, calibration, and safety reasoning. Even when models predict the correct action, 37-58% of cases still yield incorrect safety judgments, and no model passes more than 2 of 5 diagnostic checks. The dataset is publicly available on Hugging Face.
https://arxiv.org/abs/2607.05264
Motor impairments, including tremor, bradykinesia, gait abnormalities, and postural instability, are common across many neurological and movement-related conditions. Conventional clinical assessments are often intermittent and may fail to capture subtle temporal variations in motor behavior. While wearable IMUs and third-person video have shown promise for objective motor assessment, third-person recordings raise privacy concerns and require constrained acquisition setups. In contrast, egocentric vision provides a more naturalistic and privacyaware alternative. In this work, we introduce EgoInertia-MI, a multimodal benchmark dataset combining synchronized egocentric video and wearable IMU signals for motor impairment analysis. The dataset contains 19 upper- and lower-body activities performed by healthy volunteers simulating varying levels of motor impairment severity levels: no impairment, mild impairment, and severe impairment. We establish two benchmark tasks: action recognition and motor impairment severity estimation, and evaluate multiple unimodal and multimodal baselines. Experimental results show that egocentric video provides strong cues for motor impairment assessment, while multimodal fusion achieves the best overall performance, reaching 0.78 Macro-F1 for severity estimation and 0.93 Macro-F1 for action recognition. These findings highlight the potential of combining egocentric vision and wearable sensing for ecologically valid and privacy-aware motor assessment. Code and data are available at:this https URL.
https://arxiv.org/abs/2607.03934
Skeleton-based action recognition models have recently shown strong performance on large-scale benchmarks with general actions. However, directly transferring them to domain-specific tasks e.g., healthcare monitoring, is often suboptimal, as such tasks are narrow in scope and may be relevant to only a subset of general motion priors. Moreover, not all pretrained motion patterns are equally useful for a specific task, and retaining less relevant components may hinder adaptation and increase computational cost. To address these challenges, we propose Prior-Adaptive Transfer of Skeletons (PATS), a framework that adapts general skeleton-based models by selectively retaining task-relevant motion priors while filtering redundant ones during transfer. PATS follows a standard pipeline that extracts skeleton signals from videos and employs a spatio-temporal backbone pre-trained on general actions. The key contribution lies in a novel Adaptive Prior Transfer module, which performs model compression as a prior selection mechanism through iterative pruning and refinement. Experiments on two specific action recognition tasks, Alzheimer's detection and fall detection, show consistent improvements in both performance and efficiency over competitive baselines. The code will be released upon acceptance.
https://arxiv.org/abs/2607.03327
Modern video surveillance systems generate far more video streams than human operators can effectively monitor, making automated analysis essential for timely detection of security events. This paper presents a unified multi-task deep learning framework that simultaneously performs face recognition with zone-based authorization, automatic license plate recognition, weapon detection, fire and smoke detection, and human action recognition on a shared GPU platform. Among the integrated modules, two task-specific deep-learning models are proposed in this work to address scenarios that are insufficiently represented in publicly available datasets: a single-class weapon detector fine-tuned on a merged and relabeled dataset, achieving a mean average precision (mAP@0.5) of 0.947, and a SlowFast-R50 action recognition model trained on a purpose-built vandalism dataset comprising 614 video clips, achieving 94.33% classification accuracy. To improve robustness in continuous video, all detection modules are integrated into a temporal event-validation architecture based on multi-frame confirmation, confidence-weighted voting, and cascaded filtering, transforming frame-level predictions into reliable security events. Each module is evaluated independently on established public datasets (LFW, D-Fire, FIRESENSE, and UCF-Crime), followed by integrated end-to-end system evaluation. The proposed temporal validation strategy reduces the fire and smoke false-alarm rate from 52% to 4% and improves video license plate exact-match accuracy from 66.7% to 81.8%, while the complete framework maintains real-time operation with a per-frame latency below 100 ms on commodity hardware. These results demonstrate that combining specialized deep-learning models with temporal event validation provides an effective and practical solution for reliable real-time intelligent video surveillance.
https://arxiv.org/abs/2607.03131
Video Large Language Models (VideoLLMs) have shown strong progress in video understanding, yet they still suffer from hallucinations that are inconsistent with visual evidence. Existing benchmarks mainly focus on object hallucination or coarse action perception, leaving a key video-specific problem underexplored: motion hallucination, in which models infer human motions that are absent from the video. We present MoHallBench, a benchmark for diagnosing motion hallucination in VideoLLMs. MoHallBench systematically evaluates three major sources of hallucination: co-occurrence priors, sequential inference, and similarity confusion. It contains 11,306 video clips and 40,493 question-answer pairs, covering binary-choice, multiple-choice, and generative settings. We further introduce a bi-directional questioning protocol with bias-aware metrics to reduce affirmation bias in binary evaluation. Experiments on ten recent open-source VideoLLMs reveal a clear decoupling between action recognition and hallucination resistance, as models that perform well on positive action recognition often fail on adversarial negatives. Among all settings, sequential inference hallucination is the most severe, showing that current models tend to over-infer expected outcomes from partial motion cues. Our analyses further confirm that stronger priors and finer-grained similarity substantially amplify hallucination. We hope MoHallBench can facilitate future evaluation and mitigation of motion hallucination in VideoLLMs.
https://arxiv.org/abs/2607.01117
Skeleton-based action recognition has achieved remarkable success by exploiting joint coordinates and their topological connections, yet prevailing methods overwhelmingly assume complete and clean skeleton inputs. In real-world deployments, such as egocentric vision, crowded surveillance, wearable devices, or edge robotics, limited field-of-view (FoV) frequently causes substantial joint visibility dropout, leading to severe performance degradation that existing models are largely unprepared to handle. To bridge this critical yet underexplored gap, we introduce PartialVisGraph, a novel hypergraph framework tailored for robust skeleton action recognition under constrained FoV. We first construct highly expressive hypergraphs by introducing learnable virtual hyperedges that form a soft incidence matrix, capturing flexible high-order dependencies beyond conventional pairwise graphs. We then propose the Single-Head Sample-Adaptive Transformer, which adaptively aggregates joint features onto hyperedges while explicitly incorporating a visibility prior. This prior selectively gates information flow, preventing occluded or out-of-view joints from corrupting reliable feature propagation. We further establish rigorous evaluation protocols with realistic FoV simulation benchmarks on NTU RGB+D 60 and 120. Extensive experiments demonstrate that PartialVisGraph consistently achieves state-of-the-art accuracy under partial visibility, with gains of up to 68.8\% on subsets with severe FoV restrictions compared to recent strong baselines, while remaining superior on full-visibility settings. Our approach offers a principled and practical pathway toward deployable skeleton-based action understanding in unconstrained environments.
https://arxiv.org/abs/2607.00716
Effective multi-task learning for surgical scene understanding is fundamentally hindered by annotation granularity mismatch; temporal workflow tasks such as phase recognition, step recognition and anticipation benefit from dense frame-level supervision, whereas pixel-level spatial tasks including instrument segmentation and action recognition are only sparsely annotated on selected keyframes due to prohibitive labeling costs. This supervision imbalance undermines shared representation learning and limits joint optimization across heterogeneous surgical tasks. To address this, we propose Flow-guided Annotation for Robust Operating Scenes (FAROS), a flow-guided label interpolation framework, that combines zero-shot segmentation-based mask propagation with optical flow estimation to overcome the limitations of appearance-based propagation under challenging surgical conditions such as occlusion, smoke, and motion blur, generating temporally consistent dense pseudo labels from sparse keyframe annotations. The densified instrument masks and action labels are integrated into a unified Transformer-based multi-task framework that jointly learns surgical phase recognition, step recognition, anticipation, instrument segmentation, and action recognition, enabling balanced optimization between dense temporal supervision and sparse spatial supervision. The label interpolation quality of FAROS is first validated on the DAVIS 2017 benchmark under a sparse ground-truth protocol, confirming robust propagation beyond the surgical domain. Extensive experiments on GraSP, MISAW, and AutoLaparo benchmarks further demonstrate that FAROS significantly improves cross-task representation learning and enhances holistic surgical scene understanding performance across spatio-temporal tasks.
https://arxiv.org/abs/2606.26634
Given the widespread prevalence of depression and its consequential impact on individuals and society, it is crucial to obtain objective measures for early diagnosis and intervention. As a multidisciplinary topic, these objective measures should be interpretable and accessible to health care professionals, ensuring effective collaboration and treatment planning in the realm of mental health care. Even though current automated depression diagnosis approaches improved over the last decade, a critical gap exists as they often lack affect-specificity and interpretability, limiting their practical application and potential impact on mental health care. In particular, interpretability from temporal activities from videos when deep models are used is not fully explored. In this study, we present a novel framework for analyzing Deep Neural Networks' decisions when trained on facial videos, specifically focusing on automatic depression severity diagnosis. By fine-tuning Deep Convolutional Neural Networks (DCNN) pre-trained on Action Recognition datasets on depression severity facial videos from AVEC depression dataset, our framework is able to interpret the model's saliency maps by examining face regions and temporal expression semantics. Our approach generates both visual and quantitative explanations for the model's decisions, providing greater insight into its reasoning. In addition to this interpretability, our video-based modeling has improved upon previous single-face benchmarks for visual depression diagnosis, resulting in enhanced predictive performance. Overall, our work demonstrates the successful development of a framework capable of generating hypotheses from a facial model's decisions while simultaneously improving depression's predictive capabilities.
https://arxiv.org/abs/2606.25606
Children are naturally energetic, and during their spontaneous activities, they often encounter potentially dangerous situations, especially when lacking parental supervision. Identifying actions that pose risks plays a crucial role in ensuring their safety. This paper build a novel challenging dataset, namely KidRisk, including 2,500 short videos of children's actions and 10,000 images for dangerous action of children. We also introduce a benchmark on our newly constructs dataset and find that traditional deep learning models demonstrated limited effectiveness on these tasks. Therefore, we develop vision-language based baselines with exceptional context understanding of visual information. Our proposed methods achieved an accuracy of 83.53% in classifying children's actions and 96.14% in recognizing children's dangerous actions, significantly outperforming traditional approaches. These results confirm that vision-language models are not only feasible but also highly effective in detecting hazardous actions, contributing positively to safeguarding children's safety.
https://arxiv.org/abs/2606.25298
The incorporation of additional modalities into action recognition models increases their performance across a wide range of settings. However, how this additional information can contribute to making the models more robust remains underexplored, particularly for the case of multi-modal out-of-distribution (OOD) detection. While methods exist that regularize the multi-modal training process with OOD detection in mind, they still apply off-the-shelf OOD detectors designed for the uni-modal case during inference, discarding important information. Based on an interesting relationship we find between the multi-modal and uni-modal predictions, we propose to use this signal to build a post-hoc detector explicitly designed for the multi-modal scenario. We combine this new source of information with a feature-space score, which detects off-manifold samples in the multi-modal space, and normalize them by the multi-modal logits. In doing so, the proposed hybrid detector is compatible with existing training-time approaches and consistently improves performance. Experiments on a wide range of established datasets from the MultiOOD benchmark show that, on average, our approach outperforms the state of the art. Our results show the importance of explicitly considering the different modalities at inference time for multi-modal OOD detection.
https://arxiv.org/abs/2606.24404
While point cloud foundation models have significantly advanced point cloud video understanding, existing parameter-efficient fine-tuning (PEFT) methods still suffer from two critical limitations: prohibitive annotation costs for large-scale point cloud datasets and severe memory bottlenecks. In this paper, we aim to mine richer supervision signals from existing data rather than blindly scaling datasets. A further key principle is that the memory footprint of fine-tuning must be drastically reduced compared to full fine-tuning, which remains elusive for current PEFT techniques. Driven by these challenges, we identify three core desiderata: data-, parameter-, and memory efficiency, and present PoinTriE, a unified framework that excels along all three dimensions. For pre-training, pseudo-motion trajectories are synthesized via rigid transformations, paired with text corpora and 2D projections derived from raw point clouds. We then propose a Geometric-Motion Duality Network optimized via multimodal contrastive learning, rigid rotation prediction, and motion distribution divergence to produce dense self-supervision. During fine-tuning, we freeze the pretrained backbone and only update a lightweight Spatio-temporal Side Network built with LoRA units. Equipped with a gradient flow masking strategy, PoinTriE simultaneously reduces memory consumption and parameter overhead. Extensive experiments confirm that PoinTriE establishes new state-of-the-art results on action recognition and semantic segmentation tasks.
https://arxiv.org/abs/2606.24175
Egocentric video understanding is inherently limited by the narrow perspective of wearable cameras: a single viewpoint, a single modality, a single model cannot capture the full richness of human action. We argue that a truly expressive egocentric representation must subsume complementary knowledge across viewpoints, modalities, and foundation model representations, yet remain deployable from egocentric video alone. To this end, we introduce a hierarchical multi-teacher distillation framework that produces UNIEGO, a unified egocentric encoder trained with nine teachers spanning ego-exo viewpoints, RGB, depth, and skeleton modalities, and four foundation models. Rather than distilling directly from heterogeneous teachers whose incompatible architectures and feature geometries induce conflicting gradients, our framework interposes a layer of representation-specific Proxy models that translate diverse teacher knowledge into a homogeneous egocentric space. A second distillation stage, Selective Proxy Distillation (SPD), then adaptively selects, for each training sample, the subset of proxies that are both correct and confident, distilling exclusively from reliable supervision and suppressing erroneous signals. SPD is further stabilized by initializing UNIEGO as a learned convex combination of proxy parameters, placing the unified model in a well-conditioned region of the loss landscape before distillation begins. UNIEGO achieves state-of-the-art performance across three egocentric video understanding tasks - action recognition, video retrieval, and action segmentation on three challenging ego-exo benchmarks, outperforming naive multi-teacher distillation baselines and demonstrating that structured, proxy-mediated knowledge transfer yields richer and more discriminative egocentric representations.
https://arxiv.org/abs/2606.20559
Human Action Recognition (HAR) is frequently investigated in Human-Robot Collaboration (HRC) research to understand what actions have been performed and hence the state of a collaborative task. Accurately tracking an assembly state from HAR is however not fully investigated, and in realistic scenarios is not a trivial task. This research systematically investigates and compares methods for tracking assembly state using action recognition inputs. Investigations using two diverse datasets and five state tracking approaches, including logic-based, Hidden Markov Model (HMM), and neural network (NN) methods, show that optimal approaches are not uniform across different tasks and that different methods fail under different circumstances. Testing is performed using both simulated inputs with varying noise levels and realistic inputs from a HAR model. Results show NN and HMM methods can perform well in tasks with limited variability, but for other scenarios logic-based approaches can be more robust. Methods which model expected action duration are also important for tasks with repeated actions where no additional sensing is provided.
https://arxiv.org/abs/2606.20150
Fine-grained action recognition in egocentric video is challenging for Vision-Language Models (VLMs): actions often differ only in small visual cues, and a single model tends to be biased toward a subset of these cues. We propose Divide, Deliberate, Decide, a fully-local, zero-shot multi-agent framework in which (i) a VLM orchestrator chunks the video and proposes a top-k candidate label list per segment, (ii) an ensemble of heterogeneous VLM specialists, drawn from different open model families, engages in a structured deliberation that includes a peer-consultation round of questions, and (iii) agent rankings are aggregated with a Borda count and the orchestrator re-ranks its own prediction in light of the specialists' evidence. The entire pipeline runs locally with no fine-tuning. Experiments show that our method positively improves zero-shot action recognition performance over the baseline, highlighting the influence of a heterogeneous deliberation step, showing that the gain stems from decorrelated model priors rather than from additional compute.
https://arxiv.org/abs/2606.17627
Translating video demonstrations into executable robot commands remains challenging because existing methods often fail to identify which objects are functionally involved in the demonstrated action. As a result, they may generate commands that are linguistically plausible but operationally ambiguous. We propose an object-centric video understanding framework that decouples action recognition from object identification to generate precise, grammar-free manipulation commands. Our approach integrates Temporal Shift Modules (TSM) for efficient spatio-temporal action classification with a novel \textbf{Object Selection} algorithm that identifies task-relevant objects through trajectory-based role classification, blur detection, and overlap minimization. The selected objects are then processed by Vision-Language Models (VLMs) for robust category recognition and zero-shot generalization. Evaluated on a modified Something-Something V2 dataset, our method achieves 86.79\% action classification accuracy and BLEU-4 scores of 0.337 on standard objects and 0.261 on novel objects. These results improve over the strongest task-specific baseline by 80.2\% and 143.9\%, respectively. Larger gains are observed in METEOR and CIDEr, reaching 157.9\% and 171.7\% on novel objects. Across all semantic metrics, our approach consistently outperforms task-specific methods and remains competitive with, or surpasses, large general-purpose VLMs while retaining a modular, object-centric design.
https://arxiv.org/abs/2606.16470