Pet-ReID-IMAG

The 3rd place solution to CVPR2022 Biometrics Workshop Pet Biometric Challenge

Introduction

• 😊 We only trained one model (ResNeSt) with different scales (i.e., 224, 256, and 288), respectivel achieved 91.7% and 86.27% in phase A and B.
• 🚀 Traing time cost ~1.5 hour with a V100 16GB, so easy, no bells and whistles!
• 👀 Techical details are described in our PDF.
• 👉 The train/test data can be obtained from 百度云, Google drive.
• 👉 The weights can be obtained from 百度云, Google drive.
• Click on the star ⭐, Thank you ❤️

Requirements

• PyTorch 1.7.0+cu101
• torchvision 0.8.1+cu101

Prepare data

cd ./Pet-ReID-IMAG
mkidr data

# Download train_dir.zip  
unzip train_dir.zip  

# move train_dir  to ./pet_ReID-IMAG/data

Training instruction

pip install -r  requirements.txt; cd fastreid/evaluation/rank_cylib; make all

bash train1.sh
bash train2.sh
bash train3.sh
bash train4.sh

Test on Pet Challenge

bash predict.sh

Acknowledgement

A large portion of code is borrowed from fast-reid, many thanks 👍 to their wonderful work!

Thanks to my teammate Zijun Huang for his great support 😊!