search

happynear / AMSoftmax

A simple yet effective loss function for face verification.
MIT License
487 stars 130 forks source link

The way of using AMSoftmax #16

Closed Anguliachao closed 6 years ago

Anguliachao commented 6 years ago

Dear AMSoftmax team, Thanks for the great work, I've checked out your paper and prepare to try my owndata on the repo, however I'm a bit confused, would you mind telling me the relation between Amsoftmax and Sphereface? I noticed AMSoftmax seems to be the latest result on your experiment, but there's no many instructions of manipulation . As far as I recognized , should the only difference between Sphereface and AMSoftmax be the prototxt file? besides I can not only obtain AMSoftmax repo, but to keep Sphereface repo and follow the steps to train?

happynear commented 6 years ago

Yes, follow the steps of SphereFace to train but using the Caffe and prototxt files provided in this repo.

Anguliachao commented 6 years ago

Many thanks !

Anguliachao commented 6 years ago

Yes, follow the steps of SphereFace to train but using the Caffe and prototxt files provided in this repo.

@happynear Sorry to bother again, I successfully ran evaluation code on sphereface network, the result also turns out pretty good, however when I borrow the code to test AMSoftmax, specify the AMSoftmax caffemodel and prototxt file , the accuracy turns out only 60%, Is there any tricks I didn't notice or maybe this evaluation code is not suitable on AMSoftmax, would you kindly telling me how can I obtain the results in the paper? Thanks a lot. below is my testing AMSoftmax result:

`fold    ACC
----------------
1       60.67%
2       61.17%
3       63.83%
4       60.33%
5       62.00%
6       57.83%
7       61.00%
8       58.50%
9       62.00%
10      64.67%
----------------
AVE     61.20%`
happynear commented 6 years ago

face_deploy_20.zip Try this deploy file?