02
Computer vision · Evaluation
A score still needs a decision.
An interactive study of similarity scores: how changing an acceptance threshold changes false matches and missed matches.
Compare 0.82 with 0.73. One more actual match gets through. So do two non-matches.
ACCEPTANCE THRESHOLD
Where would you set the threshold?
PAIRSIMILARITY SCOREDECISION
01
Reject · score 0.52 · actual non-match
02
Reject · score 0.58 · actual non-match
03
Reject · score 0.63 · actual non-match
04
Reject · score 0.69 · actual match
05
Reject · score 0.73 · actual non-match
06
Reject · score 0.78 · actual match
07
Reject · score 0.81 · actual non-match
08
Accept · score 0.85 · actual match
09
Accept · score 0.89 · actual match
10
Accept · score 0.92 · actual match
11
Accept · score 0.96 · actual match
12
Accept · score 0.98 · actual match
Actual match Actual non-match Threshold
5/ 7Matches accepted
0False accepts
2Matches missed
No false accepts here. Some real matches are still missed.
A threshold simulation with made-up scores and labels. The interface calculates outcomes; no vision model runs here.