analiza aiqbench
This study evaluates the transferability of frozen V-JEPA2 predictive video representations to heterogeneous sensor suites, including radar and depth, for classifying breathing activity and human pose. The research provides a controlled analysis of fusion strategies, revealing that simple mean pooling performs comparably to complex query-based methods under balanced supervision while confirming that specific loss functions like Consensus-JEPA offer no additional benefit in this context.
Learning Multimodal Sensor Fusion with JEPA Representations for Breathing Activity and Pose Recognition
arXiv:2610.15891v1 Announce Type: cross Abstract: Predictive video representations do not by themselves specify how a multimodal system should handle missing sensors. We transfer frozen V-JEPA2 features to a set of heterogeneous sensors comprised by RGB video, depth video, and radar time series. A shared fusion model uses modality queries, fixed sensor slots, and explicit availability to classify breathing activity and pose from all seven non-empty sensor subsets. The contribution is a controlled evaluation of subset supervision, pooling capacity, and representation transfer on 39 OMuSense-23 subjects with five subject-group folds and three seeds. With matched downstream parameter counts, the final model reaches recording-level activity macro-F1 of $75.23%\pm0.60$ averaged across seven subsets. Supervising all subsets in each batch improves the matched query-pooling control by 1.923 percentage points (conditional 95% subject-bootstrap interval [0.647, 3.209]), while query pooling shows no clear advantage over matched mean pooling under balanced supervision. Radar probes support transfer beyond pooled wavelet pixels, 64-token inference approximately preserves mean activity performance, and the tested Consensus-JEPA loss adds no measured benefit against its exact control.
Abstract:Predictive video representations do not by themselves specify how a multimodal system should handle missing sensors. We transfer frozen V-JEPA2 features to a set of heterogeneous sensors comprised by RGB video, depth video, and radar time series. A shared fusion model uses modality queries, fixed sensor slots, and explicit availability to classify breathing activity and pose from all seven non-empty sensor subsets. The contribution is a controlled evaluation of subset supervision, pooling capacity, and representation transfer on 39 OMuSense-23 subjects with five subject-group folds and three seeds. With matched downstream parameter counts, the final model reaches recording-level activity macro-F1 of $75.23%\pm0.60$ averaged across seven subsets. Supervising all subsets in each batch improves the matched query-pooling control by 1.923 percentage points (conditional 95% subject-bootstrap interval [0.647, 3.209]), while query pooling shows no clear advantage over matched mean pooling under balanced supervision. Radar probes support transfer beyond pooled wavelet pixels, 64-token inference approximately preserves mean activity performance, and the tested Consensus-JEPA loss adds no measured benefit against its exact control.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2610.15891 [cs.CV]
(or arXiv:2610.15891v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2610.15891 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 268 cuvinte
Subiecte conexe
PRIMS: Physics-guided Representation for Fluid Identification in Multimodal Sensing
arXiv · 28 iul. 2026
EMMI: Edge Multi-Modal Intelligence for Communication-Efficient MLLM Inference via Fused Representation Compression
arXiv · 11 sept. 2026
An Attention-based Model for Robust Forecasting with Missing Modality
arXiv · 16 iun. 2026
Towards robust multimodal 3D object detection via visual foundation models
arXiv · 22 sept. 2026