资源算法RangeLoss

RangeLoss

2019-09-19 | |  116 |   0 |   0

RangeLoss For Gluon

My implement of Range Loss for Deep Face Recognition with Long-tail using MxNet/Gluon

Note

  • To simplify the problem,the train data that are fed into network can't be shuffled. For example, if your train data of a mini-batch contain 2 classes each with 3 examples, your label must something like this [1,1,1,4,4,4]. In order to do that, I implement a simple dataloader called RangeLossDataLoader,you can find it in DataLoader.py.

  • I also test the impact of whether to normalize the output features, the normalized features times a constant value(40 in my test) to scale the norm of features.

  • due to my careless, the plot of features training without range loss was plotted on train set while the features training with range loss was plotted on test set. So don't be surprised that the features training with out range loss look better.

Image

SoftMax without normalize features

SoftMax with normalize features

Range Loss without normalize features

Range Loss with normalize features


上一篇:Range Loss

下一篇:fast-style-transfer

用户评价
全部评价

热门资源

  • TensorFlow-Course

    This repository aims to provide simple and read...

  • seetafaceJNI

    项目介绍 基于中科院seetaface2进行封装的JAVA...

  • mxnet_VanillaCNN

    This is a mxnet implementation of the Vanilla C...

  • DuReader_QANet_BiDAF

    Machine Reading Comprehension on DuReader Usin...

  • Klukshu-Sockeye-...

    KLUKSHU SOCKEYE PROJECTS 2016 This repositor...