| # Two-Stage Segmentation of Pelvic Bone Fragments with Injuries in CT/X-ray Images Using nnUNet | |
| Here is the code for the solution from the SMILE team that ranked 2nd in Task1 and 1st in Task2 of the PENWIN Challenge. | |
| (https://pengwin.grand-challenge.org/result/) | |
| The models used in the two tasks were all trained using nnunetv2 with default configuration. | |
| ## Installation | |
| 1. Clone the repository: | |
| ```bash | |
| git clone https://github.com/yuepeiyan/PENGWIN_Challenge.git | |
| 2. Install nnunetv2: | |
| ```bash | |
| cd nnUNet | |
| pip install -e . | |
| 3. Download the CT weights and place the `models` folder into the `Inference/CT` directory. (Using Task 1 as an example, the steps for Task 2 are identical.) | |
| ## Usage | |
| 1. Copy an image to the `test/input/images/plevic-fracture-ct` directory. | |
| 2. Run `test_run.sh` to build and run a docker container that performs inference on the image placed in the folder mentioned above. | |
| 3. The output will be saved to the `test/output/images/plevic-fracture-ct-segmentation` directory. | |
| If you prefer not to run a Docker container, you can simply use the `inference_one_image` function in `two_stage_inference.py`. | |
| This function allows you to perform inference on a single image and save the output to a specified path. | |