wangbyz / label-smoothing-regularization-caffe

caffe implementation of Label Smoothing regularization described in "Rethinking the Inception Architecture for Computer Vision

caffe implementation of label-smoothing regularization.

Bad News: This method harms the performance in my experiments in face recognition field about 2%.

Label-smoothing regularization（LSR）

A mechanism to regularize the classifier layer by estimating the marginalized effect of label-dropout during training.

Usage

Reference

Rethinking the Inception Architecture for Computer Vision

caffe

Thanks to all the contributors of caffe and Rethinking the Inception Architecture for Computer Vision  

Others 

Data augmentation

There are 7 types of data augmentation (shift, zoomImg, rotateImg, modHSV, modRGB, JpegCompression , smoothFilter)

And shift、zoomImg and rotateImg are so important due to the unstability of face alignment.

Note:

1. You should carefully read the code in src/caffe/layers/image_data_transform.hpp before you use those data augmentation.
2. JpegCompression and smoothFilter harm the performance in face recognition in my experiments。
3. Take care of parameters in image_data_transform.hpp（line 13 to 17） because those settings may be just suitable for the face images of my experiments(112X96 images same as sphereface)

Usage