Face recognition confidence scores all seem very high #2
Description
Hi,
I have implemented your api into a database application I am working on and functionality-wise does everything beautifully! Thanks for creating such a great system. However, I am having trouble with the results of the face recognition with all 10 returned candidates having very similar confidence scores. Sometimes the face being recognised is not even the first one.
Example: 0.9995, 0.9994, 0.9994, 0.9993, 0.9993, 0.9993, 0.9992, 0.9992, 0.9992, 0.9992
Interestingly, when I run your test script "test-integration", the first time after a fresh install it works without failure, but when run subsequently it fails to detect Elon and Trump.
I am also getting this in the run log every time a face is detected:
"2025-12-13 11:32:56.858 node[2674:27271] 2025-12-13 11:32:56.841911 [W:onnxruntime:, execution_frame.cc:874 VerifyOutputSizes] Expected shape from model of {1,16800,2} does not match actual shape of {1,29126,2} for output 837"
I am running on a Mac M4 chip and OS: 15.6.1
Any help would be very welcome.