can not perform same as your pretrained res101

[ruiming46zrm]

hello

Evaluation: LFW Acc: 0.9968333333333332, CFP_FP Acc: 0.8959999999999999, AgeDB Acc: 0.9726666666666667

it's my training result as 20 epoch , hyper param as you set .. acc on cfp-fp is quite bad...

[HuangYG123]

train on which dataset? I use refined ms1mv2 provided by the authors of arcface, which has 85742 ids.

[ruiming46zrm]

@HuangYG123 in deepglint(trillion pairs), 18w id there. batchsize300. arcface train with deepglint is better than ms1mv2.

[ainnn]

@HuangYG123 in deepglint(trillion pairs), 18w id there. batchsize300. arcface train with deepglint is better than ms1mv2.

Hey, what are the results on ms1mv2 and deepglint with arcface? thx.

[HeshamAMH]

please ruiming46zrm, I test the model provided by the author on LFW but the result is 0.661 not 99.+. So, please I need a link to LFW and its pair file. I think the problem for me is LFW itself.

Thanks in advance

[vietth-bka]

@HuangYG123 I got the results here when using file pretrained model CurricularFace_Backbone.pth. These are far more different from the pubic results. Please check your file CurricularFace_Backbone.pth
Evaluation: LFW Acc: 0.661, CPLFW Acc: 0.5271666666666667

Thanks

[luhairong11]

@HuangYG123 I got the results here when using file pretrained model CurricularFace_Backbone.pth. These are far more different from the pubic results. Please check your file CurricularFace_Backbone.pth
Evaluation: LFW Acc: 0.661, CPLFW Acc: 0.5271666666666667

Thanks

I have the same problem, are you sloved?

[marcohuber]

@luhairong11
I downloaded the (preprocessed) LFW data including the pairs file from here and achieved at least on LFW = 99.783 using the provided CurricularFace model. Still a bit worse, but I guess it's the minor differences in preprocessing. So the provided pre-trained model shouldn't be the problem, maybe there is something wrong with your used preprocessing/pairs?

[luhairong11]

@luhairong11 I downloaded the (preprocessed) LFW data including the pairs file from here and achieved at least on LFW = 99.783 using the provided CurricularFace model. Still a bit worse, but I guess it's the minor differences in preprocessing. So the provided pre-trained model shouldn't be the problem, maybe there is something wrong with your used preprocessing/pairs?

I downloaded the preprocessed LFW data is https://github.com/ZhaoJ9014/face.evoLVe,it is according to author provide https://github.com/HuangYG123/CurricularFace in Usage,is it wrong?

[michellerybak]

@HuangYG123 I got the results here when using file pretrained model CurricularFace_Backbone.pth. These are far more different from the pubic results. Please check your file CurricularFace_Backbone.pth
Evaluation: LFW Acc: 0.661, CPLFW Acc: 0.5271666666666667
Thanks

I have the same problem, are you sloved?

This happens when you don't load the model checkpoint properly!

[billpsomas]

@michellerybak what do u mean by loading the model checkpoint correctly?

[michellerybak]

Sorry, should've specified!

In config.py make sure you specify the correct path for MODEL_ROOT = , otherwise it just loads the model without the pretrained weights and it does not throw an error.

[billpsomas]

Sorry, should've specified!

In config.py make sure you specify the correct path for MODEL_ROOT = , otherwise it just loads the model without the pretrained weights and it does not throw an error.

@michellerybak perfect, you were right. Actually, I also added the path to BACKBONE_RESUME_ROOT = ... and it worked fine. I think this was the problem. I am able to reproduce the numbers now! Thank you!

[michellerybak]

Sorry, should've specified!
In config.py make sure you specify the correct path for MODEL_ROOT = , otherwise it just loads the model without the pretrained weights and it does not throw an error.

@michellerybak perfect, you were right. Actually, I also added the path to BACKBONE_RESUME_ROOT = ... and it worked fine. I think this was the problem. I am able to reproduce the numbers now! Thank you!

ah, I think that's probably what I meant and just misspoke. I'm glad you got it working!