deeplabc

BehaviorScope-X integrates with pose-estimation model-based behavior-classification workflow for multi-animal videos. A pose checkpoint is reused for both keypoint detection and frozen visual feature extraction, while a temporal classifier learns behavior from visual tokens, pose-self features, and inter-animal relational geometry.