This repo is an open source entry to the 2023 Scroll Prize aka The Vesuvius Challenge. The name for this repo comes from Venus, who was the patron saint of Pompeii, the city covered in Ash by the volcano Vesuvius.
Dependencies:
- Segment Anything
- HyperOpt
- PyTorch
- Tensorboard
pip install -r requirements.txt
To run a hyperparameter sweep use:
python sweep.py
To run a saved model use the evaluation notebook eval.ipynb. Follow the copy paste instructions to submit to Kaggle.
This repo was built live on YouTube, you can find the playlist here:
Various sources used for reference:
- PyTorch Pre-trained Models
- Pretrained ViT Multiscale Vision Transformers
- Pretrained Video Transformer (SWIN)
- A ConvNet for the 2020s
- Adapting Pre-trained Vision Transformers from 2D to 3D through Weight Inflation Improves Medical Image Segmentation
- Segment Anything
@misc{ashenvenus-vesuvius-challenge-2023,
title={AshenVenus: Open Source Entry to the 2023 Scroll Prize},
author={Hugo Ponte},
year={2023},
url={https://github.com/hu-po/ashenvenus}
}