Contribution to the MVSEG Challenge (MICCAI 2023) by Marek Wodzinski (2nd place).
The challenge website: Link
Here you can see the full source code used to train / test the proposed solution.
Only the final experiment is left (the one used for the final Docker submission).
- In order to reproduce the experiment you should:
- Download the MVSEG dataset Link
- Update the hpc_paths.py and paths.py files.
- Run the parse_mvseg.py
- Run the run_aug_mvseg.py
- Run the training using run_segmentation_trainer.py
- And finally use the trained model for inference using inference.py
The network was trained using HPC infrastructure (PLGRID). Therefore the .slurm scripts are omitted for clarity.
Please cite the MVSEG (TODO) if you found the source code useful.
Please find the method description: Description