Signs of progress?! + Papyrology team Q&A

The race for the First Letters prize is heating up 📜🔥

It’s always hard to tell with a single letter — is it real or is it hallucination? But FlyingMeatBall posted an interesting discovery on Discord (plus code on Github): they ran conventional (non-ML) techniques on an area segmented by our recently assembled segmentation team, and found what looks like a psi (ψ) in plain sight:

We do have to be careful in jumping to conclusions, since it’s easy for us humans to see patterns where there are none, but this discovery has a lot going for it: the letter has roughly the right size and orientation, and lines up with the papyrus fibers.

The full segment is quite large, so we encourage everyone to look at this (and other) segments and see if you can find anything. In particular, if you’re competing in the Ink Detection prize on Kaggle, try to run your algorithms on one of the many surface volumes we have available on our data server. For example, this segment can be found at http://dl.ash2txt.org/full-scrolls/Scroll1.volpkg/paths/20230509182749/layers/ (sign up to get the login details).

We have now segmented almost 200 cm², so there’s a good chance that your existing Kaggle algorithm works on some of it.

Here are your “3 easy steps to win $40,000” if you have a strong Kaggle algorithm.. 😁

1. Run your algorithm on one of the many segments. Do keep in mind that the resolution here is 8µm instead of the 4µm of the fragments, so you might have to retrain or somehow adjust your algorithm.

2. Can you find some letters? Minimum 10 letters in a 4 cm^2 area (one pixel = 8µm, so roughly a 7000x7000 pixel area).

3. If you find some letters, submit for the First Letters prize. The first to make an eligible submission wins $40k, the second wins $10k! Profit!

Of course, it might not be as easy as 1-2-3 — there are different resolutions, image intensities, and so on — but there’s a good chance you can make rapid progress and claim this prize!

Papyrology Team Q&A

Thursday May 25th at 8am PST we’ll have a Q&A with part of our Papyrology Team. We’re collecting questions for them in the #papyrology-qa channel on Discord. Also see the Livestream page.

Community news

• We have added lots more information to the website:

  • An overview of the current state of the competition, including lots of links to software.

  • Lots more information about all the data on the server.

  • Links to advanced tutorials.

• The new “Optical Flow” algorithm for Volume Cartographer by RICHI (on Discord) has landed! Our contract segmenters have been using it for a while, and have been quite productive with it.

• The Quick Segment folks have added a tutorial inside the tool.

  

• Brett Olsen made another Kaggle notebook with efficient zarr loading and caching, generalized from his and Moshe Levy’s implementation of data loading in VolumeAnnotate.

• Santiago Pelufo has been experimenting with finding strands of papyrus in 3D space.

  

• Luke Farritor’s scroll viewer clearly shows the 3d-printed casing which the scrolls were held into during scanning.

  

• WayneWayneHello is continuing to work on their own papyrus scans.

  

• Way more on the Kaggle forums and Discord!