CCUP: A Controllable Synthetic Data Generation Pipeline for Pretraining Cloth-Changing Person Re-Identification Models

Due to the high cost of constructing Cloth-changing person reidentification (CC-ReID) data, the existing data-driven models are hard to train efficiently on limited data, which causes the issue of overfitting. To address this challenge, we propose a low-cost and efficient pipeline specific to CC-ReID tasks for generating controllable and high-quality synthetic data simulating the surveillance scenarios. Particularly, we construct a new self-annotated CC-ReID dataset named Cloth-Changing Unreal Person (CCUP), containing 6,000 IDs, 1,179,976 images, 100 cameras, and 26.5 outfits per individual. Based on this large-scale dataset, we introduce an effective and scalable pretrain-finetune framework for enhancing the generalization of the traditional CC-ReID models. The extensive experimental results demonstrate that our framework could improve the original models such as two typical models TransReID and FIRe² after pretraining on CCUP and finetuning on a benchmark, and outperform other state-of-the-art models. The dataset is available at: https://github.com/yjzhao1019/CCUP.