xERM.PC.public

Cross-Domain Empirical Risk Minimization for Unbiased Long-tailed Classification

• This codebase is built on PC.

We currently provide the training code of xERM-PC for CIFAR100-LT and ImageNet-LT.

Install

To obtain the same results, please make sure to set up the same environment.

 conda create  --name xERM --file spec-list.txt

Preliminaries

• Prepare dataset: CIFAR-100, ImageNet-LT
  • Please download those datasets following Decoupling.

CIFAR100-LT

Imbalance Ratio: 100

Step 1: Train PC model

python main.py --seed 1 --cfg config/CIFAR100_LT/ce100.yaml --gpu 0

Step 2: Train xERM-PC model:

python main.py --seed 1 --cfg config/CIFAR100_LT/ce100.yaml --gpu 0 --xERM

Step 3: Evaluate:

python main.py --seed 1 --cfg config/CIFAR100_LT/ce100.yaml --gpu 0 --xERM --test

Imbalance Ratio: 50

Step 1: Train PC model:

python main.py --seed 1 --gpu 0 --cfg config/CIFAR100_LT/ce50.yaml

Step 2: Train xERM-PC model:

python main.py --seed 1 --gpu 0 --cfg config/CIFAR100_LT/ce50.yaml --xERM

Step 3: Evaluate:

python main.py --seed 1 --gpu 0 --cfg config/CIFAR100_LT/ce50.yaml --xERM --test

Imbalace Ratio: 10

Step 1: Train PC model

python main.py --seed 1 --gpu 0 --cfg config/CIFAR100_LT/ce10.yaml

Step 2: Train xERM-PC model:

python main.py --seed 1 --gpu 0 --cfg config/CIFAR100_LT/ce10.yaml --xERM

Step 3: Evaluate:

python main.py --seed 1 --gpu 0 --cfg config/CIFAR100_LT/ce10.yaml --xERM --test

ImageNet-LT

Step 1: Train PC model

python main.py --cfg  config/ImageNet_LT/ce.yaml --seed 1  --gpu 0,1,2,3

Step 2: Train xERM-PC model:

python main.py --cfg  config/ImageNet_LT/ce.yaml --seed 1  --gpu 0,1,2,3 --xERM

Step 3: Evaluate

python main.py --cfg  config/ImageNet_LT/ce.yaml --seed 1  --gpu 0,1,2,3 --xERM --test

*We modify the original config of PC. To run the original config of PC, please change the config/ImageNet_LT/ce.yaml to config/ImageNet_LT/ce_pc.yaml

Citation

if you find our codes helpful, please cite our paper:

@inproceedings{beierxERM,
  title={Cross-Domain Empirical Risk Minimization for Unbiased Long-tailed Classification},
  author={Zhu, Beier and Niu, Yulei and Hua, Xian-Sheng and Zhang, Hanwang},
  booktitle={AAAI Conference on Artificial Intelligence},
  year={2022}
}

License

The use of this software is released under BSD-3.