mouxiaohuang/dul

PyTorch implementation of DUL (Data Uncertainty Learning in Face Recognition, CVPR2020)

About Me

• 👋 I currently work as an Algorithm Engineer at Huawei Noah's Ark Lab, where I specialize in Deep Learning and Computer Vision. I'm passionate about developing robust algorithms capable of effectively addressing practical, everyday challenges.
• Please refer to https://mouxiaohuang.com for more information.
• Don't hesitate to reach out! You can contact me at huangmouxiao[at]gmail[dot]com.

dul's Issues

About softmax loss of cls loss

It seems like you did not use softmax loss to classification as the paper said.

This class

You just used a fully connected layer, so it is another CEloss.

Is there anything that I have missed? or it is a little mistake?

pretrained models

Hi,

First thanks for your great work.

I am wondering if you feel comfortable uploading pretrained models.

Looking forward to your reply.
Thanks

关于代码的问题

您好我想问一下，这个模型在测试阶段还需要使用方差吗？在测试阶段应该直接用mu就行了吧，但是我看代码好像是都使用了

关于标准差的范围

你好。原文中提到将σ限制到[0,1]范围内，以保证KL损失在单调递减范围内，但是我在代码里面好像没有看到有限制std的范围，请问是什么原因？

Checkpoints

Hi,
Are checkpoints available for this model or should I train the model to get them?
Thanks.

KL-loss is too large

Thanks for your great work!
I noticed the KL-loss in your log is about 4. But when I run this code, the KL-loss was too large and turned out to be as high as 400 at the beginning! It's not normal but I haven't figure it out. Do you know the possible reason? Thanks a lot!