Contrastive Clothing and Pose Generation for Cloth-Changing Person Re-Identification

Vuong D. Nguyen, Pranav Mantini, Shishir K. Shah; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 7541-7549

Cloth-Changing Person Re-Identification (CCRe-ID) aims at matching an individual across cameras after a long period of time presenting variations in clothing compounded with changes in pose viewpoint etc. In this work we propose CCPG: Contrastive Clothing and Pose Generation framework for CCRe-ID. Beyond appearance CCPG captures cloth-invariant body shape information using a Relational Graph Attention Network. Training a robust CCRe-ID model requires a large range of clothing variations and expensive cloth labeling which is lacked in current CCRe-ID datasets. To address this we propose a GAN-based model for clothing and pose transfer across identities to augment images of wider clothing variations and of different persons wearing similar clothing. The augmented batch of images serve as inputs to our proposed Fine-grained Contrastive Losses which not only supervise the Re-ID model to learn discriminative person embeddings under long-term scenarios but also ensure in-distribution data generation. Results on CCRe-ID datasets demonstrate the effectiveness of our CCPG framework. Code will be available at https://anonymous.4open.science/r/CCPG-ReID.