上一条: From Pixel to Patch: Synthesize Context-Aware Features for Zero-Shot Semantic Segmentation
下一条: Exploiting motion information from unlabeled videos for static image action recognition