Back to the Challenge: $100K Kaggle Surface Detection

Find papyrus surfaces in 3D CT, and help unlock the Herculaneum library.

Vesuvius Challenge exists to read the carbonized Herculaneum scrolls. We’re returning to our roots with a new Kaggle competition and a $100,000 prize pool.

This is what a sample from the dataset looks like: binary labels (in red) are overlaid on the 3D CT scan data.

The task (briefly):

• Problem: Detect the papyrus surface inside 3D CT scans.

• Data: CT chunks + binary masks (smoothed sheet positions).

• Goal: Train a model that reproduces those labels on held‑out data.

• Scoring: A topology‑aware linear blend rewarding both voxel accuracy and surface connectivity (no gaps, holes, sheet‑switches or mergers).

• Why it matters: Topology‑clean surfaces feed our virtual unwrapping pipeline! With better surface detection we can unwrap (and read) the scrolls!

Make history. Climb the leaderboard. Help read the scrolls.

Join the competition!