Mining False Positive Examples for Text-Based Person Re-Identification

Text-based person re-identification (ReID) aims to identify images of the targeted person from a large-scale person image database according to a given textual description. Most existing methods focus on establishing instance-level and word-region-level correspondences. However, these methods generally rely heavily on the similarity contributed by matched word-region pairs while neglecting the decisive mismatched word-region pairs, resulting in false-positive matching. Accordingly, we propose to mine false-positive examples (MFPE) via a jointly optimized multi-branch architecture. Specifically, we delicately design a cross-relu loss to amplify the dissimilarity between matched and mismatched word-region pairs. Furthermore, we propose a false-positive mining (FPM) branch to highlight the role of mismatched word-region pairs, effectively preventing false-positive matching. Extensive experiments on CUHK-PEDES demonstrate the superior effectiveness of MFPE. Our code is released at https://github.com/xx-adeline/MFPE.