Learning relation-based features for fine-grained image retrieval

Yingying Zhu, Gang Cao, Zhanyuan Yang, Xiufan Lu

Abstract: Highlights•A relation extraction module that models implicit relationships by expectation-maximization.•A relation-based convolutional descriptor aggregation method for global descriptors.•A novel pipeline for weakly supervised fine-grained image retrieval.•Extensive experiments show that the proposed method achieves new state-of-the-art.