schema_version string | experiment string | preregistered_contract dict | sources dict | generation dict | inference dict | runtime dict | cells list | elapsed_seconds float64 |
|---|---|---|---|---|---|---|---|---|
1.0 | finite-thickness fusion pilot | {
"source_comment": "https://github.com/ScrollPrize/villa/issues/191#issuecomment-5162354182",
"output_keys": [
"prob",
"gt_surface",
"turn_id"
],
"probability_dtype": "float16",
"array_order": "ZYX",
"thresholds_for_downstream_readout": [
0.4,
0.5,
0.6
],
"primary_readout": "fus... | {
"painter_pr": "https://github.com/Diego-dcv/vesuvius-topological-grid/pull/2",
"painter_commit": "2c483dd16ebfdcd35740864a5e1e30d1fcc03af9",
"painter_sha256": "9d69cad386f34e9ba3b11c54514e1396ed4b5dcd47c8ad44eb75f9c2bb57eccf",
"villa_commit": "4d5c9e605df86eb5060c0753faadc8b2aba44175",
"harness_sha256": "54... | {
"pitches_um": [
300,
250,
200,
160
],
"seeds": [
0,
1
],
"papyrus": 90,
"noise_sigma": 6,
"sheet_um": 150,
"voxel_um": 30,
"z_window_mm": 8,
"columns": 12
} | {
"patch_size": 192,
"step_fraction": 0.5,
"foreground_channel": 1,
"blending": "separable Gaussian, sigma=patch/8, minimum weight 1e-4",
"test_time_augmentation": false,
"parameter_count": 102349770
} | {
"python": "3.12.13 (main, Mar 4 2026, 09:23:07) [GCC 11.4.0]",
"platform": "Linux-6.12.90+-x86_64-with-glibc2.35",
"torch": "2.5.1+cu121",
"cuda_device": "Tesla P100-PCIE-16GB"
} | [
{
"tag": "seed0_pitch300",
"file": "seed0_pitch300.npz",
"file_sha256": "85b914d4a87062b3621b7983e54ce125c4d71dfbf18a6233c5e49f4b0debcabb",
"seed": 0,
"pitch_um": 300,
"shape_zyx": [
266,
497,
929
],
"output_bytes": 220462046,
"patches": 90,
"inference_secon... | 1,236.504233 |
Vesuvius finite-thickness fusion pilot
This is the preregistered eight-cell handoff requested by Jinho Jeong in ScrollPrize/villa issue #191. It is designed to measure whether a frozen Vesuvius surface checkpoint keeps neighbouring finite-thickness sheets separated as their true air gap closes.
Fixed design
- painter:
2c483dd - checkpoint:
scrollprize/surface_recto_059_redo,Model_epoch499.pth - papyrus level 90; noise sigma 6
- fixed 150 µm sheets; 30 µm voxels; 12 columns; 8 mm Z window
- pitches 300, 250, 200, and 160 µm
- independent generation seeds 0 and 1
- frozen 192³ tiled inference at 50% step, no test-time augmentation
Each NPZ contains exactly three arrays in ZYX order:
| Key | Dtype | Meaning |
|---|---|---|
prob |
float16 |
Frozen checkpoint foreground probability |
gt_surface |
uint8 |
Exact finite-thickness material mask |
turn_id |
int16 |
1-based winding-turn instance; 0 is air |
The preregistered downstream readout is fusion rate over neighbour-sheet sites
and false-split rate over single-sheet controls at thresholds 0.4, 0.5, and
0.6. Recall and near-sheet AUC in manifest.json are secondary descriptive
metrics, not the primary endpoint.
Integrity
manifest.json pins the painter, Villa revision, inference harness,
checkpoint, generated input volumes, output arrays, and NPZ files by SHA-256.
Its SHA-256 is:
506f3353ba834bcdb839cf4000d2ec6b1a18e074ce1fb9c10dce24696e7299dd
The local independent validator reported ALL CHECKS PASSED: 156.
python validate_outputs.py .
Scope
These are analytic synthetic volumes with exact instance labels. They support a controlled fusion-versus-gap measurement; they do not estimate accuracy on real scroll data and do not establish that any predicted sheet is scientifically correct.
- Downloads last month
- 160