Reproduce “Unsupervised Pre-training for Person Re-identification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dengpan Fu, Dongdong Chen, Jianmin Bao, Hao Yang, Lu Yuan, Lei Zhang, Houqiang Li, Dong Chen
Researchers from USTC and Microsoft Research addressed person re-identification challenges by introducing LUPerson, a massive unlabeled dataset with over 4 million images, and pioneering large-scale unsupervised pre-training. This approach significantly boosts both supervised and unsupervised Re-ID methods, particularly yielding substantial performance gains in data-scarce environments.
Key insights
- Large-scale unsupervised pre-training on domain-specific unlabeled data (LUPerson) effectively learns generic and transferable person Re-ID features, outperforming ImageNet pre-training.
- For fine-grained tasks like Re-ID, color-distorting data augmentations are detrimental, while stronger Random Erasing is highly beneficial for learning robust features.
- A smaller temperature parameter (e.g., 0.07) in the contrastive loss is optimal for person Re-ID, enhancing discriminability between subtle inter-class differences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.