上一条： Beyond without forgetting: Multi-task learning for classification with disjoint datasets

下一条： Activity Image-to-Video Retrieval by Disentangling Appearance and Motion