Learning Using Privileged Information for Zero-Shot Action Recognition

Zero-Shot Action Recognition (ZSAR) aims to recognize video actions that have never been seen during training. Most existing methods assume a shared semantic space between and unseen intend directly learn mapping from visual the space. This approach has challenged by gap paper presents novel method uses object semantics as privileged information narrow and, hence, effectively, assist learning. In particular, simple hallucination network is proposed implicitly extract testing without explicitly extracting objects cross-attention module developed augment feature with semantics. Experiments on Olympic Sports, HMDB51 UCF101 datasets shown outperforms state-of-the-art large margin.