Medical-Segmentation-Wavelet-Transform

A repo of a project in DS503 - Spring 2024 in KAIST.

Installation

conda env create -f environment.yml
conda activate Wavelet_MedSeg

Run the Code

# Run the Wavelet model for segmentation #
python train.py --num_epoch [# of epochs] --save_content --batch_size [# of batch size] --config [# name of the config file]

### example ### 
python train.py --num_epoch 100 --save_content --batch_size 14 --config monuseg_wavelet_NoGAN.yml 

# Run the Unet model for segmentation # 
python train_Unet.py --num_epoch [# of epochs] --save_content --batch_size [# of batch size] --config monuseg.yml

Inference

# Run the Wavelet model for segmentation #
python infer.py --config [# name of the config file]

# Run the Unet model for segmentation # 
python infer_Unet.py --config monuseg.yml

### The output of inferencing is located in the experment folder ###

BF-Score

# Change the root folder is the file that contain the inference output images in the file 'bf-score.py' then run #

python bf-score.py

Dataset

Data setlink:

• Monuseg in this project
• Original of Monuseg

Please download then change the data path in the config file.

Citation

This wavelet idea comes from Wavelet Diffusion. If you use this wavelet idea, please cite below:

@inproceedings{phung2023wavelet,
  title={Wavelet diffusion models are fast and scalable image generators},
  author={Phung, Hao and Dao, Quan and Tran, Anh},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={10199--10208},
  year={2023}
}