Adaptive Semantic Consistency for Cross-domain Few-shot Classification

This is the official code for https://arxiv.org/pdf/2308.00727.

If you find this repo useful for your research, please consider citing this paper

@article{lu2023adaptive,
  title={Adaptive Semantic Consistency for Cross-domain Few-shot Classification},
  author={Lu, Hengchu and Shao, Yuanjie and Wang, Xiang and Gao, Changxin},
  journal={arXiv preprint arXiv:2308.00727},
  year={2023}
}

Dataset & Pretrained Mdoel Download

Please follow here to download dataset and pretrained model

Running

• To run contrastive finetuning on EuroSAT data (default target domain), run python Supcon_ASC.py, with 5-way 5-shot as default setting

Acknowlegements

Our code is based on On the Importance of Distractors for Few-Shot Classification.