This repository contains the code used to achieve the 34th place in the Vesuvius Challenge - Ink Detection competition, hosted on Kaggle. You can visit the competition page here.

This section provides information about how to download the necessary datasets from Kaggle.

This dataset is essential for the main task. It can be downloaded from the following Kaggle competition page:

• Vesuvius Challenge Ink Detection

The trained weights for the inference model can be downloaded from:

• VCID Trained Weight

The datasets for training can be downloaded from:

• VCID validation
• VCID modified inklabels
• VCID tile iamges 256
• VCID tile iamges 1024

1. Clone this repository:

git clone --recursive https://github.com/tmyok/kaggle-vesuvius-challenge-ink-detection.git

2. Download the datasets into the input directory. Refer to Data Acquisition for more information.

3. If necessary, run a Docker container with the following command:

sh docker_container.sh

Ensure that your hardware has at least a total of 40 GB of GPU RAM and then run the following command:

python3 train.py model={unet3d, residualunetse3d, mit_b2} fold={0,1,2,3,4}

If your hardware has less GPU RAM, adjust the resize_ratio in the configuration file (located at ./working/configs/model/{UNet3D, ResidualUNetSE3D, mit_b2}.yaml).

The trained weights will be exported to ./output/{UNet3D, ResidualUNetSE3D, mit_b2}.

To perform validation, run the following command:

python3 validation.py fold={0,1,2,3,4}

The results will be exported to ./output/validation.

To perform inference, run the following command:

python3 inference.py --image_path ../input/vesuvius-challenge-ink-detection/test/a/surface_volume/ --mask_path ../input/vesuvius-challenge-ink-detection//test/a/mask.png

The results will be exported to ./output/inference.