AI RESEARCH

Purify-then-Align: Towards Robust Human Sensing under Modality Missing with Knowledge Distillation from Noisy Multimodal Teacher

arXiv CS.CV

ArXi:2604.05584v1 Announce Type: new Robust multimodal human sensing must overcome the critical challenge of missing modalities. Two principal barriers are the Representation Gap between heterogeneous data and the Contamination Effect from low-quality modalities. These barriers are causally linked, as the corruption