Real‑time multi‑person action recognition from 2D pose (keypoints) using YOLO, ViTpose, Kalman Filter and custom temporal GraphSAGE and Transformers.
computer-vision pytorch action-recognition skeleton-based-action-recognition graphsage real-time-detection keypoint-estimation temporal-transformer temporal-graph-networks pygeometric multiperson-tracking
-
Updated
Mar 26, 2026 - Python