Cross-Level Distillation and Feature Denoising for Cross-Domain Few-Shot Classification

Code implementation of ICLR 2023 paper entitled "Cross-Level Distillation and Feature Denoising for Cross-Domain Few-Shot Classification"

English version:

Updates on 2023-05-23

We firstly release the code of training with Cross-Level Distillation (CLD)

Train the model with the command:

python oho.py --dir tmp/dir_you_want --target_dataset EuroSAT --target_subset_split splits/EuroSAT_unlabeled_20.csv --alpha 2 --bsize 32 --coef 1

BTW, if you want to Train the SimCLR baseline, just modify the line 260 in methods/student9.py by removing the distillation loss. We will release the formal code soon!

TODO on 2023-05-23

Releasing the code of testing stage, including Feature Denoising (FD). We are really sorry that we are still sorting. If you are in a hurry, you can also contact me in advance by sending emails to me. You are always welcome to contact us!!!

---

中文版：

2023年5月23日更新

我们首先开源了训练Cross-Level Distillation (CLD)的代码

你只需要：

python oho.py --dir tmp/dir_you_want --target_dataset EuroSAT --target_subset_split splits/EuroSAT_unlabeled_20.csv --alpha 2 --bsize 32 --coef 1

如果你想训练SimCLR baseline的话，只需要修改methods/student9.py中第260行，把蒸馏loss去掉就可以了。后续我们也会开放正式的代码！！！

2023年5月23日TODO

开放测试阶段的代码，包括Feature Denoising (FD). 我很抱歉我们还在整理！如果你着急的话，也可以给我发邮件。随时欢迎交流沟通！！！