PAct official inference was attempted on laptop_real.jpg using a manual SAM2 merge:
- base = segments [0, 8, 14, 15]
- screen = segment [4]
The official pipeline loaded and sampled sparse structure + SLat, but failed in articulation decode:
ValueError: cannot convert float NaN to integer
Location: modules/utils/articulation_utils.py -> parse_tree()
This page keeps the preprocessing visuals and records the failure honestly.