https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/overview
- Docker: 23.0.1
- rocker: 0.2.10
- off-your-rocker
Build a docker image.
make build-docker
pipx install git+https://github.com/sloretz/off-your-rocker --include-deps
Download data from the competition page.
$ tree /path_to_input -L 2
/path_to_input
└── vesuvius-challenge-ink-detection
├── sample_submission.csv
├── test
└── train
Download pre-trained models of 3D-ResNet.
gdown 12FxrQY2hX-bINbmSrN9q2Z5zJguJhy6C
gdown 1K9oiny9ENYODFxjFBTKdGeoDvOzPu4qQ
mkdir checkpoints
mv r3d18_KM_200ep.pth checkpoints
mv r3d34_KM_200ep.pth checkpoints
make run-docker LOCAL_DATA_PATH=/path_to_input
python -m inkdet.tools.preprocess
train <CONFIG_PATH>
Run shell scripts in the scripts
directory to train models for the submission.