Backfill Misses failing

[adjablonski]

Bug description

Hi,

I'm trying to generate LAD estimates using Phytograph - I appreciate the quick turn around on the import error. I have no issues importing scans, however I am now encountering an issue with generating the backfill misses needed for the LAD calculation. If I use a .las file, the backfill misses process stalls at 85% and never returns a message saying the reconstruction failed. I then tried an .e57 file, which then returned the error below.

I'm not sure if this issue is caused by my data or something Phytograph is doing. The point cloud is exported from RiScan Pro with basic processing (filtering and scan registration). The example point cloud (.las and .e57) and relevant scan metadata can be found in the dropbox link below.

Thanks again!

https://www.dropbox.com/scl/fo/0hspzm476dfknnyw9ap4e/AOLwJp3PkK4Vl0kcz59HJPs?rlkey=ownv35soxc14pvf4yjlitr0ij&st=16tmal61&dl=0

"Backfill Misses failed for ExampleScan_01.e57
Could not reconstruct sky/miss points for this scan: gapfillLiDARMisses): bad allocation. The scan grid is too sparse or irregular to gap-fill. If it carries a timestamp, ensure each return keeps it; otherwise re-import a format that retains misses (E57 / structured PLY)."

Environment

• Phytograph: 0.74.0
• Backend: 0.74.0
• PyHelios: 2f90c00
• Helios (C++): 62152d3
• OS: win32

Session logs

phytograph-logs-2026-08-31T18-43-46.txt

[bnbailey-psl]

Thanks again for reporting this. I've been able to reproduce everything and fix it (and thanks for sending the actual input files so hypotheses could be confirmed against your actual data).

The LAS hang was caused by the fact that your VZ-600i instrument sweeps the mirror downward relative to the timestamp increase (everything I've worked with so far is the opposite), which resulted in a sign error that caused an infinite loop. That was a silly one, should have caught that.

The E57 threw bad allocation on a different path: the row/column fill creates one point per empty grid cell with no cap, and your 95.8M-cell grid needed ~8–10 GB as a single contiguous block on Windows. On Mac and Linux that can overflow into swap, but not for Windows it apparently has to be contiguous. So it's now chunked to keep the memory requirement lower.

Unfortunately, I couldn't find a work-around, except for me to backfill misses and send it to you. Here is your scan with misses added
here. I'm not sure if you will still run into memory issues because it's a ton of data with misses. I'm working on reducing the memory footprint of misses. Also you might run into a limitation with exporting of LAD data, which I've also fixed.