$12,000 Ink Detection Followup Prize (28 Aug)
Let's run the Ink Detection models on the scrolls!!
With the Ink Detection winners announced, the world now has ten high-performance open source models that can detect ink in CT scans of Herculaneum papyrus. It’s time to put them to work!
Today we’re announcing the Ink Detection Followup Prize, which will be awarded to open source contributors who advance the state of the art in open source ink detection model in a way that brings us all closer to reading the full scrolls.
The 10 winning teams have now all published their code, and made writeups and video presentations, all of which is linked below.
Details:
Six prizes total: two $3,000 prizes, and four $1,500 prizes
Deadline: Submissions are closed on Monday August 28th 11:59pm PT.
Any tools, documentation, notebooks, analysis, that use the code from the Ink Detection Prize. Must be open source under a permissive license (MIT, Apache).
Ideas:
Simply running a winning Ink Detection submission on the main scroll! Maybe it just works?? (If so, be sure to also submit for the $50,000 First Letters Prize!)
Easy to run notebook / Docker container / VM image that can run all models.
Greyscale outputs instead of binary classification.
Ensemble model of all the winning submissions.
Adapting or fine-tuning models so they work better on the full scrolls (e.g. to account for the 4µm vs 8µm voxel size difference).
Analysis that shows how to change our segmentation techniques to get better results with Ink Detection models.
Modify models to be trained on (and output) the infrared photos instead of hand-labeled binary classification.
Analysis that characterizes the distribution shift between the fragment data and the scroll data (on top of their differing voxel sizes) using visualization, quantification, or other methods, and approaches to bridge this domain gap
You may make multiple submissions. We may award a single prize to a set of multiple submissions by one person or team.
Your submission will be judged subjectively: the Technical Review Team will decide what they think are the best open source contributions.
If you are submitting as a team, the team leader should make the submission, and is responsible for distributing the prize money among the team.
Submit using this form.
Modification to Segmentation Tooling 2
We’re also making a slight modification to the Segmentation Tooling 2 prize: you have to also make a submission by Monday August 28th 11:59pm PT, though the prize won’t be judged until the original date of September 15th. This ensures that you have time to get feedback from us and the segmenters, and avoids last-minute submissions that don’t get used by the segmenters.
Winning teams
1st place: ryches
Writeup Github Inference notebook
2nd place: RTX23090
Writeup Github Inference notebook
3rd place: wuyu
Writeup Github Inference notebook
4th place: POSCO DX - Heeyoung Ahn
Writeup Github Inference notebook
5th place: Aksell
Writeup Github Inference notebook
6th place: chumajin
Writeup Github Inference notebook
7th place: OverthINKingSegmenter
Writeup Github (inference in repo)
8th place: Luck is all you need
Writeup Github (inference in repo)
9th place: still 1 fold, 2 net
Writeup Github Inference notebook
10th place: Feng Qilong
Writeup Github Inference notebook