CAER-skeleton

A new skeleton-based affective dataset with seven emotion categories and joint-confidence annotations for fine-grained emotion recognition. Most available skeleton-based emotion datasets remain limited to four basic categories and lack joint-level reliability annotations, constraining the development of fine-grained affective measurement models. To promote research in this direction, we establish a seven-category emotion-labeled skeleton action dataset and incorporate confidence scores for all detected joints, supporting both emotion recognition and reliability modeling.