deimsdeutsch
commented
4 years ago @jaisanant0 I am also training vargfacenet on faces_glint dataset. Have you checked the accuracy on megaface by running the script ?
https://github.com/deepinsight/insightface#test-on-megaface
msg me on dimsdeutsch@gmail.com if you would like to collaborate.
I am training vargfacenet with faces_glint dataset, below are my result
testing verification.. (12000, 512) infer time 10.401622999999994 [lfw][177500]XNorm: 23.159211 [lfw][177500]Accuracy-Flip: 0.99700+-0.00314
testing verification.. (14000, 512) infer time 11.917347999999999 [cfp_fp][177500]XNorm: 21.214161 [cfp_fp][177500]Accuracy-Flip: 0.91829+-0.01338
testing verification.. (12000, 512) infer time 10.231041000000001 [agedb_30][177500]XNorm: 23.092572 [agedb_30][177500]Accuracy-Flip: 0.96883+-0.00837
The accuracy on verification datasets looks good, but when i manually check the similarity between two different faces the similarity score comes higher than 50%.
Can anyone explain me why this is happening?
@nttstar