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

when tensorflow implementation release #1

Closed emma-sjwang closed 6 years ago

emma-sjwang commented 6 years ago

Hello! Thanks for your wonderful work and sharing. Just wonder when will release the tensorflow version of L-GM loss.

Thanks a lot.

WeitaoVan commented 6 years ago

Hi, Emma

For the tensorflow code, I have tested the classification task only on small datasets such as CIFAR100.

In fact the implementation in Tensorflow is straightforward. Because you don't have to write back-propagation. You can write the loss according to the formulation. For convenience and better convergence, you can first try fixing the covariance matrix to Identity Matrix. :) I will release the tensorflow code very soon.

Best wishes.

WeitaoVan commented 6 years ago

Hi, it is released now.