Skip to content

Unveiling the secrets of an ancient library buried by Mount Vesuvius, this Kaggle competition, supported by the Vesuvius Challenge organization, tasked participants with detecting ink from 3D X-ray scans of charred scrolls preserved in a Roman villa in Herculaneum.

License

Notifications You must be signed in to change notification settings

association-rosia/vesuvius-challenge

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

📜 Vesuvius Challenge - Ink Detection

This exciting competition, available on Kaggle and supported by the Vesuvius Challenge organization, aims to bring back to life an ancient library buried under the ashes of a volcano. In this competition, our mission was to detect ink from 3D X-ray scanners. Thousands of scrolls were part of a library located in a Roman villa in Herculaneum. This villa was buried by the eruption of Mount Vesuvius nearly 2000 years ago. The scrolls were charred by the heat of the volcano and are now impossible to open without damaging them.

🏆 Challenge ranking

The score of the challenge was the F0.5 score.
Our solution was in the top 10% (out of 1249 teams) with a F0.5 score equal to 0.620813 🎉.

The podium:
🥇 ryches - 0.682693
🥈 RTX23090 - 0.682443
🥉 wuyu - 0.681137

🛠️ Data processing

Tilling Method

Ink labels with tilling method performing on it

In order to process the ultra-high definition images effectively, we employed an image tiling method. The images were divided into smaller sub-images with a size of 256 by 256 pixels to feed into the model. This approach allowed us to handle the large dataset more efficiently. Additionally, we implemented a selection criterion where we only considered tiles that contained a minimum of 5% ink pixels. By focusing on these specific tiles, we were able to concentrate our efforts on the areas most likely to contain valuable information within the ancient scrolls.

🏛️ Model architecture

Architecture of EficientUnet V2 model

📝 Citing

@misc{RebergaUrgell:2023,
  Author = {Louis Reberga and Baptiste Urgell},
  Title = {Vesuvius Challenge - Ink Detection},
  Year = {2023},
  Publisher = {GitHub},
  Journal = {GitHub repository},
  Howpublished = {\url{https://github.com/association-rosia/vesuvius-challenge}}
}

🛡️ License

Project is distributed under MIT License

👨🏻‍💻 Contributors

Louis REBERGA

Baptiste URGELL

Johan MONCOUTIÉ

Releases

No releases published

Packages

No packages published