PyTorch implementation for Source-free domain adaptation with Class Prototype Discovery. This repository is based on framework from CAN and modified part of the code.
The installation can refer to CAN.
Please download datasets to experiments/datasets/ folder.
CUDA_VISIBLE_DEVICES=X python tools/train_target.py --method CPD --cfg experiments/config/config_file.yaml --weights pre_trained_source_model_weights
If you think our paper or code is helpful to you, we very much hope that you can cite our paper, thank you very much.
@article{zhou2024source,
title={Source-free domain adaptation with Class Prototype Discovery},
author={Zhou, Lihua and Li, Nianxin and Ye, Mao and Zhu, Xiatian and Tang, Song},
journal={Pattern recognition},
volume={145},
pages={109974},
year={2024},
publisher={Elsevier}
}