Open blinkbink opened 5 years ago
IMHO, I would start by checking the alignment on the faces, if it's not perfect you may want to try a different face detector
IMHO, I would start by checking the alignment on the faces, if it's not perfect you may want to try a different face detector
Ah ok, i would to try thank you
first align in RGB. Then use num_jitters value up till 5 while generating encoding. Then keep the threshold below 0.45 to avoid false positives.
Description
I'm running with cuda, tryng to compare similarity with selfie face and face on National Identity Card then got score 0.56 (under 0.60) on different people. How i can handle this ? any advices ?
i try set score 0.55 but many prediction (same people score like 0.57 or 0.58)