Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Yujian Zhao, Chengru Wu, Yinong Xu, Xuanzheng Du, Ruiyu Li, Guanglin Niu

TL;DR

Cloth-changing Re-Identification is hampered by limited labeled data across outfits. The authors propose CCUP, a large-scale, controllable synthetic CC-ReID data generation pipeline built on MakeHuman and Unreal Engine, producing 6,000 IDs, 1,179,976 images, 100 cameras, and an average of 26.5 outfits per identity. They validate a pretrain-finetune framework that pretrains TransReID and FIRe^2 on CCUP and finetunes on PRCC, VC-Clothes, NKUP, achieving state-of-the-art results and demonstrating improved cloth-irrelevant feature learning. Visualizations indicate CCUP pretraining shifts attention to body regions and key body parts, supporting improved generalization to cloth-changing scenarios. Overall, CCUP provides a cost-effective, scalable path to robust CC-ReID models and clearer avenues for future improvements in cloth-irrelevant representation learning.

Abstract

Cloth-changing person re-identification (CC-ReID), also known as Long-Term Person Re-Identification (LT-ReID) is a critical and challenging research topic in computer vision that has recently garnered significant attention. However, due to the high cost of constructing CC-ReID data, the existing data-driven models are hard to train efficiently on limited data, causing overfitting issue. To address this challenge, we propose a low-cost and efficient pipeline for generating controllable and high-quality synthetic data simulating the surveillance of real scenarios specific to the CC-ReID task. 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 capabilities of the traditional CC-ReID models. The extensive experiments demonstrate that two typical models namely TransReID and FIRe^2, when integrated into our framework, outperform other state-of-the-art models after pretraining on CCUP and finetuning on the benchmarks such as PRCC, VC-Clothes and NKUP. The CCUP is available at: https://github.com/yjzhao1019/CCUP.

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

TL;DR

Cloth-changing Re-Identification is hampered by limited labeled data across outfits. The authors propose CCUP, a large-scale, controllable synthetic CC-ReID data generation pipeline built on MakeHuman and Unreal Engine, producing 6,000 IDs, 1,179,976 images, 100 cameras, and an average of 26.5 outfits per identity. They validate a pretrain-finetune framework that pretrains TransReID and FIRe^2 on CCUP and finetunes on PRCC, VC-Clothes, NKUP, achieving state-of-the-art results and demonstrating improved cloth-irrelevant feature learning. Visualizations indicate CCUP pretraining shifts attention to body regions and key body parts, supporting improved generalization to cloth-changing scenarios. Overall, CCUP provides a cost-effective, scalable path to robust CC-ReID models and clearer avenues for future improvements in cloth-irrelevant representation learning.

Abstract

Cloth-changing person re-identification (CC-ReID), also known as Long-Term Person Re-Identification (LT-ReID) is a critical and challenging research topic in computer vision that has recently garnered significant attention. However, due to the high cost of constructing CC-ReID data, the existing data-driven models are hard to train efficiently on limited data, causing overfitting issue. To address this challenge, we propose a low-cost and efficient pipeline for generating controllable and high-quality synthetic data simulating the surveillance of real scenarios specific to the CC-ReID task. 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 capabilities of the traditional CC-ReID models. The extensive experiments demonstrate that two typical models namely TransReID and FIRe^2, when integrated into our framework, outperform other state-of-the-art models after pretraining on CCUP and finetuning on the benchmarks such as PRCC, VC-Clothes and NKUP. The CCUP is available at: https://github.com/yjzhao1019/CCUP.

Paper Structure

This paper contains 17 sections, 6 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Cloth-changing person re-identification
  4. Dataset synthesis
  5. Methodology
  6. Dataset generation
  7. Generate person skeletal mesh
  8. Simulate surveillance in scenarios
  9. Detect and label the bounding boxes
  10. Pretrain and finetune
  11. Experiment
  12. Dataset and evaluation metrics
  13. Implementation details
  14. Comparison with state-of-the-art methods
  15. Impact of different pretraining dataset
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Pipeline of our work. We first generate the skeletal mesh of person and provide a large number of clothing textures for cloth-changing. Then the skeletal mesh is imported into the three scenarios to simulate surveillance and the person in the surveillance frames is detected. Finally, we construct a large-scale dataset called CCUP containing 6,000 IDs, 100 cameras, and 1,1179,976 images. We pretrain the TransReID and FIRe$^2$ two baseline models with CCUP and finetune on PRCC, VC-Clothes, and NKUP benchmarks.
  • Figure 2: Visualization of FIRe$^2$ evaluated on PRCC. (a) Original iamges. (b) Results without pretraining. (c) Results with pretrain-finetune.