The code of paper "Step-wise Distribution Alignment Guided Style Prompt Tuning for Source-free Cross-domain Few-shot Learning"

1. Setup

conda creat --name stepspt python=3.9

conda activate stepspt

conda install pytorch torchvision -c pytorch

conda install pandas

pip install numpy

pip install argparse

pip install math

pip install os

pip install sklearn

pip install scipy

pip install PIL

pip install abc

2. Code clone

git clone https://github.com/xuhuali-mxj/StepSPT.git

cd StepSPT

3. Dataset

For the 4 datasets CropDiseases, EuroSAT, ISIC, and ChestX, we refer to the BS-CDFSL repo. For PatternNet, please get from PatternNet.

4. Run StepSPT

Based on ConvNeXt

Our method aims at improving the performance of pretrained source model on the target FSL task. We introduce the style prompt and step-wise distribution alignment, helping the pretrained large-scale model to learn the decision boundary.

Please set your data address in configs.py.

The pretrained source model 'convnext_base_22k_224.pth' can be download in ConvNeXt.

We start from run.sh. Taking 5-way 1-shot as an example, the code runing process can be done as,

python ./stepspt.py  --dtarget CropDisease  --n_shot 1

Based on other backbones

Coming soon.

5. Acknowledge

Our code is built upon the implementation of FTEM_BSR_CDFSL and TPDS. Thanks for their work.