ICLR23-BEEF

The code repository for "BEEF: Bi-Compatible Class-Incremental Learning via Energy-Based Expansion and Fusion " [paper]

@inproceedings{wang2023beef,
  title={BEEF: Bi-Compatible Class-Incremental Learning via Energy-Based Expansion and Fusion},
  author={Wang, Fu-Yun and Zhou, Da-Wei and Liu, Liu and Ye, Han-Jia and Bian, Yatao and Zhan, De-Chuan and Zhao, Peilin},
  booktitle={The Eleventh International Conference on Learning Representations},
  year={2023}
}

Prerequisites

The following packages are required to run the scripts:

• torch
• torchvision
• tqdm
• numpy

Training scripts

• Train CIFAR-100

  python main.py --config=./configs/beef/cifar-50-10.json
  

• Train ImageNet-100

  python main.py --config=./configs/beef/imagenet-50-10.json
  

• Train imbalanced protocols

  python main.py --config=./configs/beef/cifar-50-10-random.json # uniform ramdom type
  
  python main.py --config=./configs/beef/cifar-50-10-imbalance-1.json # half-half-type
  
  python main.py --config=./configs/beef/cifar-50-10-imbalance-0.9.json # exp type
  

Remember to change YOURDATAROOT into your own data root, or you will encounter errors.

Results

Experimental results show that our method achieves state-of-the-art performance.

Protocols	Reproduced Avg	Reported Avg
CIFAR-100 B50 5 steps	71.75	71.70
ImageNet-100 B50 5 steps	78.48	77.27
CIFAR-100 uniform	70.85	71.08
CIFAR-100 half-half	67.72	66.81
CIFAR-100 exp	68.86	67.85

Contact

If there are any questions, please feel free to contact the author: Fu-Yun Wang ([email protected]). Enjoy the code.