ryohachiuma / DFU-challenge

This is the 1st place solution for Diabetic Foot Ulcer challenge in MICCAI 2020.

DFU challenge repository

This repository is for DFU challenge (https://dfu2020.grand-challenge.org/evaluation/submissions/create/). This challenge will be held in MICCAI 2020 (https://www.miccai2020.org/).

This repository is based on mmdetection. Please follow the instruction guide on installation

Quick Demo

If you want to test with your image, download the one of the pretrained weights of any model (e.g. Deformable Convolution) from the link and save to the MODEL_PATH. And assume that the image is placed at IMG_PATH. And run the following command.

python demo/image_demo.py IMG_PATH configs/feet/deform_low.py MODEL_PATH 

This will visualize the bounding boxes of the wound.

Training models

Faster R-CNN, Deformable Conv2, Cascade R-CNN, pisa

python tools/train.py configs/feet/feet_low.py
python tools/train.py configs/feet/deform_low.py
python tools/train.py configs/feet/cascade.py
python tools/train.py configs/feet/pisa.py

The pretrained models can be downloaded from the link

Testing models

python demo/inference.py configs/feet/feet_low.py work_dirs/feet_low/epoch_17.pth --score-thr=0.7
python demo/inference.py configs/feet/deform_low.py work_dirs/deform_low/epoch_25.pth --score-thr=0.7
python demo/inference.py configs/feet/cascade.py work_dirs/cascade/epoch_23.pth --score-thr=0.7
python demo/inference.py configs/feet/pisa.py work_dirs/pisa/epoch_19.pth --score-thr=0.7

Merging all models

python demo/ensemble_nms.py