search

TreB1eN / InsightFace_Pytorch

Pytorch0.4.1 codes for InsightFace
MIT License
1.74k stars 423 forks source link

Do you test you model on IJB? #96

Open njtuzzy opened 5 years ago

njtuzzy commented 5 years ago

did you ever test the model on IJB? and how's the results?

TreB1eN commented 5 years ago

No，but welcome

njtuzzy commented 5 years ago

@TreB1eN, actually, I used your pretrained model SE-IR-50 and evaluation script from insightface and loose crop data from VGG group to evaluate the performance on IJBB, the results turn to be wired---it is a little bit bad as following: 0.0001-34.37%, 0.001--59.43%, 0.01--81.77%, 0.1--93.19%. do you have any comments about this results?

pansanity666 commented 3 years ago

@TreB1eN, actually, I used your pretrained model SE-IR-50 and evaluation script from insightface and loose crop data from VGG group to evaluate the performance on IJBB, the results turn to be wired---it is a little bit bad as following: 0.0001-34.37%, 0.001--59.43%, 0.01--81.77%, 0.1--93.19%. do you have any comments about this results?

hello, the evaluation code in insightface only supports keras and mxnet form of pretrained weight, how did you load in pytorch model? did you modified the code yourself? Could you please share your code, thx