Open yangxudong opened 5 years ago
I've already implemented SV-X-Softmax and it works fine!
@tiandunx Would you like to share your code to github?
@Laviyy Sorry to reply late. Of course I'm willing to share my implementation.
@tiandunx We are waiting for you! Thanks^_^
@tiandunx
any update?
Hi all, I'm so sorry for keeping you waiting. I shall release the code after ICCV. Sorry again. Thanks for keeping an eye upon this paper.
@tiandunx Would u mind tell me ur LFW accuracy? I implemented it on mnist, and the classification accuracy was not very good. As for on face recognition, the code is still running. So I cannot get the result now.
@luameows Sure, on LFW, I achieved 99.866% using MS only.
Hi all, I'm so sorry for keeping you waiting. I shall release the code after ICCV. Sorry again. Thanks for keeping an eye upon this paper.
@tiandunx After Nov 3, 2019? 😲😲😲
At the end of this month.
@tiandunx which backbone net did you use to achieve 99.866% on lfw? res-50? or attention-56?
@tiandunx
What is the value of t you used? And what is the margin loss function you used?
t = 1.2, margin loss is additive margin softmax@twmht
At the end of this month.
Is the code able to be shared? I'll be very appreciated.
The code is available now.
@tiandunx
thanks for sharing.
So you use all the default values when set is_am=True
?
def __init__(self, feat_dim, num_class, is_am, margin=0.45, mask=1.12, scale=32):
@tiandunx Hi, which backbone net did you use to achieve 99.866% on lfw? res-50? or attention-56?
why this repo is empty.