5 days left for the open source prizes + new 10x $250 private prizes
Vesuvius Challenge Newsletter #4
Open source prizes live stream
The first deadline of open source prizes is coming up next Tuesday. If you haven’t already, consider open sourcing your work and submitting it for a prize!
We’re super excited to be awarding our first prizes, so we decided to do a livestream on which we announce the winners of the 3x $2000, reveal the fun surprise we’re awarding alongside the money, and do a quick Q&A. Tune in on Friday 14 April at 11:00am PT.
Private prizes 10x $250
We're looking to learn what people are working on, and to reward folks who have privately been exploring the data. It's simple:
Tell us in 1-2 paragraphs what you have done so far and/or are planning to do (privately, we'll keep it confidential).
We select 10 submissions that get $250!
You don't need to have made much progress at all to be eligible. Even a solid idea that you haven't implemented yet could win. We will not share anything about your submission with anyone.
Email team@scrollprize.org before next Tuesday, April 11th 11:59pm PT (same as the open source prize deadline). If you have a team, please have the team leader send the email, and include who is on the team. Same details on the website.
Community news
Lots of folks have been posting their training notebooks on Kaggle.
Moshe Levy made a virtual unwrapping tool to rival our Volume Cartographer. It’s written in Python and easy to hack on!
Lev Lipinski made a cool notebook to explore seeing ink directly. Can you see the letter that appears here?
Thomas Gibbons has been using the Kaggle competition in his college class.
Brett Olsen found a simple heuristic to detect the surface of a fragment.
Kota Iizuka compared some faster Run Length Encoding algorithms for the final .csv file serialization. James Davey made a faster pixel coordinate generator. Both have been incorporated into our tutorial, which now runs a bunch faster!
Richard Carrier fantasized about what we might find in the scrolls.
Good discussion about using hand-labeled binary masks versus infrared photos for model training.
Lots more details from Seth Parker about how the scanning was done.
MatthieuFP wrote “scrollloader”, a tool for loading 3D volumes into HDF5, for more efficient processing. Similarly, Brett Olsen released a new version of his zarr-based image/volume loading tool, vesuvius_image.
hu-po and Luke Farritor have been livestreaming hacking on the Kaggle competition, and a web-based scroll viewer, respectively.
Allie Wicklund made a notebook that does some filtering of background values, to aid with segmentation.
Lots of activity on Kaggle, but still lots of room to improve!
It’s been awesome to see all the progress, and we’re excited to award our first prizes on Friday April 14th! See you then!