3DEyeSeg

:2nd_place_medal: 3D Eye Point Cloud Segmentation Challenge - OpenEDS 2021

Information about 3D Eye Point Cloud Segmentation can be found in here.

How to use

1. Use conda environment

   • Create a conda environment with python 3.7
   • Activate the environment and use install_packages.sh to install required packages
   • Modify conf/eval.yaml, checkpoint_dir in line 9 to absolute path of ckpt (in current folder)
   • Run python eval.py

2. Use docker (the method that we used)

   • Create docker image with dockerfile, for example docker build -t prj:eyeseg3d .
   • Run

docker run --gpus all --ipc=host -it --rm \
                   -v path_to_src:/mnt/Work/OpenEDS2021_EyeSeg3D \
                   -w /mnt/Work/OpenEDS2021_EyeSeg3D \
                   prj:eyeseg3d python3 eval.py

In here path_to_src is the absolute path to source code folder.

The predictions for test set will be saved to ckpt/eval/%Y-%m-%d_%H-%M-%S/viz/100/test.

Credits

Our source code is built upon on torch-points3d library.