You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, I am using torch openface for testing the performance of lfw dataset.
Since your model is caffe based, I use loadcaffe library for torch to convert the model into torch.
Standard preprocessing process of caffe is resizing the scale to [0,255] and substract the mean value of the image, and the input channel should be BGR.
However, I still cannot reproduce the performance of mobild_id model.
Can you please tell me what kind of preprocessing do you perform? Or any special operation?
Also, in the paper they claim to use Euclidean distance for face verification in LFW dataset instead of SVM or Joint Bayesian, but the github page claims that one has to use Joint Bayesian to get the 97% performance. Does this mean the provided model cannot get similar performance in the paper by using Euclidean distance?
The text was updated successfully, but these errors were encountered:
Hi, I am using torch openface for testing the performance of lfw dataset.
Since your model is caffe based, I use loadcaffe library for torch to convert the model into torch.
Standard preprocessing process of caffe is resizing the scale to [0,255] and substract the mean value of the image, and the input channel should be BGR.
However, I still cannot reproduce the performance of mobild_id model.
Can you please tell me what kind of preprocessing do you perform? Or any special operation?
Also, in the paper they claim to use Euclidean distance for face verification in LFW dataset instead of SVM or Joint Bayesian, but the github page claims that one has to use Joint Bayesian to get the 97% performance. Does this mean the provided model cannot get similar performance in the paper by using Euclidean distance?
The text was updated successfully, but these errors were encountered: