Since launching two weeks ago, there has been a whirlwind of activity. We got a lot of people competing on Kaggle, a thriving Discord community where folks analyze and discuss the data, and a ton more prize sponsors. In this newsletter, we’ll recap what prizes are currently ongoing, plus news from the community.
Prizes
We currently have the following three prizes you can compete for, with different time horizons:
April 11th 2023: open source prizes. Three $2000 prizes for publicly available contributions (code, docs, research notes, etc).
June 14th 2023: Ink Detection Progress Prize on Kaggle. $100,000 spread out over 10 prizes, for detecting ink in scroll fragments.
December 31st 2023: $700,000 Grand Prize, for reading substantial amount of texts in the intact scrolls.
We’re looking for feedback on how to best structure the remaining prize money. If you are considering to compete, but the current structure is preventing you from doing so, definitely reach out to us by email or Discord, so we can take your ideas into consideration.
Private feedback
There are lots of ways of sharing your feedback publicly, like on Discord or the Kaggle forums, but sometimes you might want to keep ideas to yourself, to keep an edge over the competition. In that case we would still love to hear from you, so we can keep tabs of what people are working on.
We can use this information to everyone’s benefit, such as prioritizing new resources to put out (e.g. data or tutorials), how to structure future progress prizes, and so on.
If you are working on something but haven’t shared it with anyone yet, please contact us by email. We’ll keep it completely confidential.
Community news
Dr. Seales published a preprint of the history of his work and the wider field.
If you’re looking for collaborators, be sure to check out these threads on Discord and Kaggle!
Brett Olsen shared a great notebook for boosting performance through Hessian denoising, as well as efficient data loading using zarr.
Good discussion on the thickness of papyrus, the internal fraying of layers inside the scrolls, and the shrinking of scrolls.
Lots of submissions on Kaggle that score around our ink-id baseline.
Interesting ideas around comparing ink detection to other problem domains.
A good clarification: the Kaggle subvolumes indeed go from inside the fragment (0) to outside the fragment / air (65).
A cool idea: detecting glue in the scrolls, to make it easier to determine where layers are. This sparked a discussion on how papyrus was made, and if you can see the “joins” of pages in detached fragments.
Benjamin Anderson published a notebook for trying the Masked Autoencoder technique on the ink detection problem.
moshelevy has been working on a custom unwrapping tool, though they have not quite shared it yet.
Similarly, lukeboi has been working on a scroll viewer.
John Costella discovered some interesting upsampling artifacts.
Speculation on what these bright spots on the “wrong side” of the Campfire Scroll are.
If you haven’t seen it already, be sure to check out last week’s Livestream Q&A, where the Vesuvius Challenge team discussed lots of great questions.
Christopher Parker wrote a wonderful article for Smithsonian Magazine.
If you need coding inspiration, there are a lot more notebooks to look at!