[Pytorch] A disentangled generative model for disease decomposition in chest X-rays via normal image synthesis
This repository implements Disentangled Generative Model(DGM) with MIMIC-CXR dataset.
See also: Paper, tensorflow implementation
We only use PA sided and clearly labeled data.
Run the shell script bellow
python data_preprocess_all.py \
--base_path <your/mimic-cxr-jpg/directory/ends/with/mimic-cxr-jpg/2.1.0/files> \
--pa_base_path <your/pa_filetered_image/directory/ends/with/physionet.org/pa_filter/pa_filtered_images> \
--csv_file </standard/test-train/split/in/mimic-cxr/mimic-cxr-2.0.0-split.csv> \
--label_csv_file </your/mimic-cxr/chexpert/label/path/mimic-cxr-2.0.0-chexpert.csv> \
--output_path </your/data/output/dir>Then, 13 files would be created in your output directory. We will use labeled_train.txt, l
53DE
abeled_validation.txt, labeled_test.txt in our training code.
Run the shell script bellow
python train.py \
--batch_size 64 \
--d_steps 1 \
--gpu <0 indexed gpu number> \
--train_file_list </your/labeled_train.txt/file> \
--val_file_list </your/labeled_validation.txt/file>You could modify run script depend on your environment.
Working..