search

WeitaoVan / L-GM-loss

Implementation of our accepted CVPR 2018 paper "Rethinking Feature Distribution for Loss Functions in Image Classification"
MIT License
172 stars 34 forks source link

question about feature var #9

Closed yuyijie1995 closed 5 years ago

yuyijie1995 commented 5 years ago

In the tensorflow readme ,you said "This is the tensorflow demo for the LGM loss. While the caffe version is more complete, this version does not support updating of the feature variances." ,but ,in the caffe edition,you also said "Fix the variances to initial values (1.0).". What is the difference?

WeitaoVan commented 5 years ago

in the readme.md please read that sentence together with the one above it

update_sigma: false

Fix the variances to initial values (1.0).

which means that the parameter update_sigma allows us to choose whether to update the variances or not.