FDG2801 / CoTrFuse

CoTrFuse

1.Download pre-trained swin transformer model (Swin-T)

• [Get pre-trained model in this link] (https://drive.google.com/drive/folders/1UC3XOoezeum0uck4KBVGa8osahs6rKUY?usp=sharing): Put pretrained Swin-T into folder "pretrained_ckpt/" and create dir 'chechpoint','test_log' in the root path.

2.Prepare data

• You can also go to https://challenge.isic-archive.com/data/#2017 to acquire the ISIC2017 dataset. Process the label from the csv file for training. Change the imgs_train_path, imgs_val_path, imgs_test_path in the train_class_after_segmentation to the path of the corresponding path.
• You can also go to https://www.kaggle.com/datasets/cf77495622971312010dd5934ee91f07ccbcfdea8e2f7778977ea8485c1914df to acquire the COVID-QU-Ex dataset.

3. Environment

• Please prepare an environment with python=3.8, and then use the command "pip install -r requirements.txt" for the dependencies.

4. Train/Test

• Run the train script on the ISIC-2017 and the COVID-QU-Ex dataset. The batch size we used is 8 and 16. If you do not have enough GPU memory, the bacth size can be reduced to 6 or 12 to save memory. For more information, contact [e-mail].

• Train

python train_ISIC.py
python train_COV.py

• Test

python test_ISIC.py
python test_COV.py

References

• TransUnet
• SwinTransformer
• Swin-Unet