Lite-MKD: A Multi-modal Knowledge Distillation Framework for Lightweight Few-shot Action Recognition.
Baolong LiuTianyi ZhengPeng ZhengDaizong LiuXiaoye QuJunyu GaoJianfeng DongXun WangPublished in: ACM Multimedia (2023)
Keyphrases
- lightweight
- multi modal
- action recognition
- video dataset
- bag of words
- activity recognition
- computer vision
- multi modality
- cross modal
- human detection
- recognition of human actions
- human actions
- uni modal
- audio visual
- image annotation
- high dimensional
- image sequences
- human activities
- video search
- video data
- wireless sensor networks
- action classification
- multiple modalities
- machine learning