A Multi-Modal Framework with Cross-Subject Pseudo-Labeling and Semantic Alignment for Micro-Gesture Recognition
Researchers have developed a new multi-modal framework for micro-gesture recognition, addressing challenges like low signal-to-noise ratio and cross-subject generalization. The system integrates skeleton joint coordinates, 3D heatmap volumes, and RGB visual features, employing a novel Cross-Modal Pseudo-Labeling strategy for domain adaptation. This approach achieved a competitive F1-score of 68.13%, securing 4th place in the 4th MiGA-IJCAI Challenge. AI
IMPACT This research advances micro-gesture recognition capabilities, potentially improving human-computer interaction and emotion detection systems.