huangzehao / caffe-vdsr

A Caffe-based implementation of very deep convolution network for image super-resolution
MIT License
273 stars 134 forks source link

This version isn't use the clip-gradients? #31

Closed Paper99 closed 7 years ago

Paper99 commented 7 years ago

I noticed that your update cancelled the clip-gradients, is it right? Why?

huangzehao commented 7 years ago

Hi, we don't need to set clip-gradients since there is not gradient explosion problem with following solver: Adam with 1e-4 learning rate.

Paper99 commented 7 years ago

Many thanks!