Segment Lens

Guide

A segment is a piece of the rolled papyrus surface that someone (or a program) traced through the X-ray CT of a scroll. It is stored as a tifxyz mesh: three images, x.tif, y.tif and z.tif, that together hold a grid of 3D points in the scan's voxels. The open data maps each segment onto several scan volumes, so one segment can have several meshes.

The colors mark what kit tifxyz check flags on the full grid:

• Holes: missing grid nodes inside the sheet (not joined to its border). The preview draws the rim around them.
• Folds: neighboring cells whose surface normals turn by more than 90 degrees, where the surface doubles back.
• Stretched: grid edges more than 1.5 times longer or shorter than the spacing the mesh's meta.json promises.

"Edges ±10%" is the share of grid edges within 10 percent of that spacing: how close the grid is to isometric, which matters for flattened renders. A clean mesh can still jump from one wrap of papyrus to the next; nothing here looks at the CT or judges ink, and none of it is an official villa or prize check.

3D previews are shown only for scans on none of the prize lists (Grand Prize, First Letters, and the title prize's PHerc. Paris 4, from kit's dated snapshot); scrolls on a list appear in the table with numbers only. Previews keep every k-th node, at most 256 per side, with each flag pooled over the block it stands for.

https://huggingface.co/datasets/SharpMeow/vesuvius-segment-atlas

# pip install huggingface_hub
from huggingface_hub import hf_hub_download, snapshot_download

path = hf_hub_download("SharpMeow/vesuvius-segment-atlas", "atlas.jsonl", repo_type="dataset")  # one file, cached
folder = snapshot_download("SharpMeow/vesuvius-segment-atlas", repo_type="dataset")

# pip install datasets
from datasets import load_dataset

default = load_dataset("SharpMeow/vesuvius-segment-atlas", "default", split="train")

# pip install pandas huggingface_hub
import pandas as pd

df = pd.read_json("hf://datasets/SharpMeow/vesuvius-segment-atlas/atlas.jsonl", lines=True)

pip install -U huggingface_hub
hf download SharpMeow/vesuvius-segment-atlas atlas.jsonl --repo-type dataset --local-dir vesuvius-segment-atlas

curl -L -O https://huggingface.co/datasets/SharpMeow/vesuvius-segment-atlas/resolve/main/atlas.jsonl

https://vesuvius-challenge-open-data.s3.us-east-1.amazonaws.com/PHerc0009B/segments/20250510172639/mesh/20250510172639-on-20250521125136-8.64um.tifxyz

# pip install numpy tifffile
import io, json, urllib.request
import numpy as np, tifffile

T = "https://vesuvius-challenge-open-data.s3.us-east-1.amazonaws.com/PHerc0009B/segments/20250510172639/mesh/20250510172639-on-20250521125136-8.64um.tifxyz"
meta = json.load(urllib.request.urlopen(T + "/meta.json"))  # scale, bbox, format
get = lambda c: tifffile.imread(io.BytesIO(urllib.request.urlopen(f"{T}/{c}.tif").read()))
xyz = np.stack([get(c) for c in "xyz"], -1)  # xyz[i, j] = (x, y, z) in scan voxels; -1 = no surface

https://sharpmeow-segment-lens.hf.space/

The whole atlas as one CSV: https://huggingface.co/datasets/SharpMeow/vesuvius-segment-atlas/resolve/main/atlas.csv
API and MCP for agents (list_scrolls, list_segments, get_segment, segment_picture): https://sharpmeow-segment-lens.hf.space/agents/
Page source: https://github.com/ChaseHendrick/Scrolls/tree/main/spaces/segment-lens

On your own machine

git clone https://github.com/ChaseHendrick/Scrolls && cd Scrolls
python -m pip install numpy tifffile imagecodecs
python -m kit atlas show PHerc0139 --top 10
python -m kit tifxyz check MESH_URL --png check.png

Credits

• Meshes: Vesuvius Challenge open data, CC BY-NC 4.0. The atlas numbers and previews carry the same license.
• This page: Apache 2.0, by Chase Hendrick (source). three.js is MIT.