← back to Trending Dw
scripts/__pycache__/heal-tile.cpython-314.pyc
66 lines
+
#WMjI � � � R t ^ RIt^ RIt^ RIt^ RIt^ RIHt ^ RIt]! R4 t ]P t^t^ t
R R ltR tR tR t]R 8X d
]! 4 R# R# )
u�
heal-tile.py — file-based seam healer for trending-dw gap candidates.
The gapgen pipeline is self-host + file-based (candidates are PNGs in /tmp/gap-cand,
no DB). wallco-ai's heal-seam-region.py does the same joint-healing but is coupled
to the all_designs DB table, so it can't be used here. This lifts its pure-numpy
heal algorithm (heal_band_mid / heal_edge_wrap) and drives it from a PNG PATH:
1. scan the tile via wallco-ai/scripts/seam-defect-boxes.py --path (get FAIL/WARN boxes)
2. heal each flagged joint in-place (v_mid/h_mid band-average, *_edge mirror-wrap)
3. save <name>_healed.png (NEVER overwrites the source — round-1 output is sacred)
4. re-scan the healed file and print before/after JSON
This is the motif-lane rescue the DTD panel required before motif generation resumes:
figural motifs (leopard/paisley/deco) tile badly (seam 15-46); healing their joints
can pull a near-miss under the <=5 gate without re-rolling (which discards the design).
Usage:
python3 scripts/heal-tile.py --path /tmp/gap-cand/TR-020__s123.png
python3 scripts/heal-tile.py --path <png> --kinds v_mid,h_mid # heal only these joints
Exit 0 if the healed tile PASSES seam (overall_max <= 5), else 1.
N)�PathzA/Users/macstudio3/Projects/wallco-ai/scripts/seam-defect-boxes.pyc �0 � V ^8� d QhR\ R\ /# )� �png�return)r �dict)�formats "�scripts/heal-tile.py�__annotate__r
# s � � M� M�d� M�t� M� c � � \ P ! \ \ \ 4 R \ V 4 .RR^<R7 pVP
^ 8w d. \
RVP P 4 R, 24 h \ P ! VP 4 # \ P d&