Human Action Recognition
- The action recognition task is often used to evaluate the quality of video features learned by self-supervised learning methods
- first trained on unlabeled video data with pretext tasks, then it is fine-tuned on action recognition datasets with human annotations to recognize the actions