xrosliang / softened-similarity-learning

Unofficial Pytorch implementation of Unsupervised Person Re-identification via Softened Similarity Learning

ARTNet Pytorch

Introduction

Unofficial Pytorch implementation of Unsupervised Person Re-identification via Softened Similarity Learning

Paper: https://arxiv.org/abs/2004.03547

Requirements

The code is written using the following environment. There isn't a strict version requirement, but deviate from the listed versions at your own risk

• python: 3.7.3
• pytorch: 1.2.0
• torchvision: 0.4.0
• matplotlib: 3.1.0
• numpy: 1.16.4
• tqdm: 4.32.1

Training

Preparing Data

1. Download and extract files for the memory table: https://drive.google.com/file/d/1Oc824mzfJ1LYRWc1R3sLWgcf_T6efpJc/view?usp=sharing
2. This implementation is built for the Market1501 dataset only. So just download it and extract to your desired location.

Configuration

1. Make a copy of config.ini
2. Edit the configurations as you see fit

Run

python train.py --config [config file path]

Testing

python test.py --config [config file path]

Extra

A pretrained model with Top-1 accuracy 0.66: https://drive.google.com/file/d/1V5JrR3Vqvj7HRQCcT99ZqBVrqUYnpRtv/view?usp=sharing