上一条： Human-centric Image Cropping with Partition-aware and Content-preserving Features
下一条： From Representation to Reasoning: Towards both Evidence and Commonsense Reasoning for Video Question-Answering
