fabianhenning / vesuvius_5

Vesuvius Challenge - Ink Detection 5th place solution

Vesuvius Challenge - Ink Detection 5th place solution

This is the training code of the 5th place solution for Kaggle Vesuvius Challenge.

Solution summary : https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/discussion/417642

Inference code : https://www.kaggle.com/code/aksell7/3dresnet18-3dresnet34-infer

HARDWARE:

Ubuntu 20.04.5 LTS

CPU AMD® Ryzen 7 3700

32GB RAM

1 x NVIDIA GeForce RTX3090

SOFTWARE: (python packages are detailed separately in requirements.txt)

Python 3.9.7

CUDA 11.7

cuddn =8.5.0.96

nvidia drivers 515.105.01

1.Place competition data to data folder as below:

/data

└── test

└── train

└── sample_submission.csv

2.Install python packages:

run install_dependencies.sh

3.Prepare data:

run prepare_data.sh

4.Train models:

run train.sh

At the end of the training, there will be 3 new folders with 6 model weights

5.Inference:

https://www.kaggle.com/code/aksell7/3dresnet18-3dresnet34-infer

In inference notebook modify chkps variable. Paths should match to model weights location.