Benchmarks for Corruption Invariant Person Re-identification

1 Nov 2021  Β·  Minghui Chen, Zhiqiang Wang, Feng Zheng Β·

When deploying person re-identification (ReID) model in safety-critical applications, it is pivotal to understanding the robustness of the model against a diverse array of image corruptions. However, current evaluations of person ReID only consider the performance on clean datasets and ignore images in various corrupted scenarios. In this work, we comprehensively establish six ReID benchmarks for learning corruption invariant representation. In the field of ReID, we are the first to conduct an exhaustive study on corruption invariant learning in single- and cross-modality datasets, including Market-1501, CUHK03, MSMT17, RegDB, SYSU-MM01. After reproducing and examining the robustness performance of 21 recent ReID methods, we have some observations: 1) transformer-based models are more robust towards corrupted images, compared with CNN-based models, 2) increasing the probability of random erasing (a commonly used augmentation method) hurts model corruption robustness, 3) cross-dataset generalization improves with corruption robustness increases. By analyzing the above observations, we propose a strong baseline on both single- and cross-modality ReID datasets which achieves improved robustness against diverse corruptions. Our codes are available on https://github.com/MinghuiChen43/CIL-ReID.

PDF Abstract

Results from the Paper


 Ranked #1 on Cross-Modal Person Re-Identification on RegDB-C (mINP (Visible to Thermal) metric)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Person Re-Identification CUHK03-C CIL (ResNet-50) mAP 16.33 # 1
mINP 22.96 # 1
Rank-1 22.96 # 1
Person Re-Identification Market-1501-C CIL (ResNet-50) mINP 1.76 # 1
mAP 28.03 # 1
Rank-1 55.57 # 1
Person Re-Identification MSMT17-C CIL (ResNet-50) mINP 0.32 # 1
mAP 15.33 # 1
Rank-1 39.79 # 1
Cross-Modal Person Re-Identification RegDB-C CIL (ResNet-50) mINP (Visible to Thermal) 38.66 # 1
mAP (Visbile to Thermal) 49.76 # 1
Rank-1 (Visible to Thermal) 52.25 # 1
mINP (Thermal to Visible) 11.94 # 1
mAP (Thermal to Visible) 47.90 # 1
Rank-1 (Thermal to Visible) 67.17 # 1
Person Re-Identification SYSU-MM01-C CIL (ResNet-50) mINP (All Search) 22.48 # 1
mAP (All Search) 35.92 # 1
Rank-1 (All Search) 36.95 # 1
mINP (Indoor Search) 43.11 # 1
mAP (Indoor Search) 48.65 # 1
Rank-1 (Indoor Search) 40.73 # 1

Methods