上一条： Memorize, Associate and Match: Embedding Enhancement via Fine-grained Alignment for Image-Text Retrieval
下一条： Hallucinating uncertain motion and future for static image action recognition
