invert_surface_volume.py

[giorgioangel]

Still to be optimized, but seems working. To be determined.

So far the script is taking as image input layers only *.png

Some multiprocessing in writing out to zarr could be beneficial.

Example command:
python invert_surface_volume.py --obj /home/giorgio/Desktop/rendering_test/20231221180251.obj --images /home/giorgio/Desktop/rendering_test/layers --output /home/giorgio/Desktop/rendering_test

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Preview	Comments	Updated (UTC)
scrollprize-org	Ignored	Preview		Aug 26, 2025 10:10am

[giorgioangel]

I modified invert_surface_volume.py and added invert_surface_volume_wrapper.py

Having a folder full of images in this format <scrollid>_<segment_id>_<layer_id>.png , one should run the wrapper as follows with --invppm_script invert_surface_volume.py and with a number of workers such that your system memory can handle it.

[giorgioangel]

After the segments renders have been inverted, one can merge them in a single per scroll zarr with the following:

invert_surface_volume_aggregator.py

Aggregate per‐segment Zarr volumes into per‐scroll Zarr volumes by taking the voxelwise maximum
across all segments of a given scroll—and optionally merging in a “predictions” Zarr—using Dask
for parallel, out‐of‐core computation.

Folder structure assumed:
segments_root/
<scroll_id>_<segment_id>/
labels/ ← a uint8 3D Zarr (DirectoryStore) with shape=(Z,Y,X)
mask/ ← a uint8 3D Zarr (DirectoryStore) with the same shape & chunks

If --predictions-zarr is provided, it should point to a directory containing
<scroll_id_predictions.zarr> stores. Each predictions Zarr must have the same overall shape
as the segment Zarrs, but may use a different chunk layout. For each scroll_id, this script
will produce:
output_root/
<scroll_id>_labels.zarr
<scroll_id>_mask.zarr

Each output Zarr has the same shape, chunk‐layout, compressor, filters, etc., as the
segment inputs, and each voxel in “labels” is the maximum over:
• all segment‐labels at that position, and
• the predictions volume (if provided, over the same voxel‐range).

Usage:
python invert_surface_volume_aggregator.py
--segments-root /path/to/segments
--output-root /path/to/aggregated_zarrs
[--predictions-zarr /path/to/predictions]
[--use-distributed]
[--n-workers N]
[--threads-per-worker T]
[--log-level LEVEL]