README.md

June 17, 2025 ยท View on GitHub

Continual Generalized Category Discovery:

Learning and Forgetting from a Bayesian Perspective

Hao Dai, Jagmohan Chauhan

University College London

University of Southampton

license

Visualization

Usage

Installation

conda create -n vbcgcd python=3.12.2
conda activate vbcgcd
pip install -r requirements.txt
mkdir datasets

Prepare Datasets

python feature_extractor/dino-cifar100.py --finetuned --output_dir datasets/cifar100

Training

python main.py --base 50 --increment 10 --pretrained_model_name dino-vitb16-sl --data_dir datasets/cifar100 --trail_name mix_increment_mngmm_dinovb16_sl_cifar_100

Citation

If you find our work useful, please cite our related paper:

# ICML 2025
@inproceedings{dai2025vbcgcd,
  title={Continual Generalized Category Discovery: Learning and Forgetting from a Bayesian Perspective},
  author={Dai, Hao and Chauhan, Jagmohan},
  booktitle={Proceedings of the 42nd International Conference on Machine Learning (ICML)},
  year={2025}
}