Hi, backward

[louielu1027]

i'm sorry, i can't find the backward code for the ghm_loss....

[libuyu]

No offense, but it seems that you are not familiar with Pytorch. The backward operation for this loss can be directly done by the Autograd mechanics (https://pytorch.org/docs/master/notes/autograd.html#autograd-mechanics), so there is no need to write a backward function here.

[louielu1027]

Thanks for your reply. You are write, i have not used Pytorch before. i want to write GHM_loss in python layer for caffe. So i need the details of the forward and the backward. And, do you have plan to write the loss for caffe?

[xialuxi]

@louielu1027 backward code : https://github.com/xialuxi/GHMLoss-caffe

[libuyu]

@louielu1027 Sorry that I am just at the AAAI conference so I could not reply in time. You can just refer to xialuxi's work. @xialuxi Good job!

[louielu1027]

@xialuxi Thanks! good job! I will try.