Vesuvius Challenge - Surface Detection

Overview

This hackathon challenges you to build a model that can virtually unwrap ancient scrolls, which are too fragile to open physically. You'll work with 3D CT scan data to train your model to trace the scroll's surface through folds and distortions. Success in this competition could help unlock lost works of philosophy, poetry, and history.

Requirements

Participants will work with 3D CT scans of carbonized Herculaneum scrolls. The goal is to detect the recto surface of the papyrus sheets, even with damage and fraying, avoiding topological errors like merging different sheets or splitting a single sheet. Submissions must be made through Kaggle Notebooks, with CPU and GPU notebooks limited to a 9-hour run-time. Internet access must be disabled, but publicly available external data and pre-trained models are allowed. Your submission should be a zip file named 'submission.zip' containing a '.tif' volume mask for each test image, matching the source image dimensions.

Prizes

The top 10 teams will receive cash prizes, with first place awarded $25,000, second place $20,000, and third place $15,000. The 4th place team will get $10,000, and teams from 5th to 10th place will each receive $5,000.