search

fqnchina / CEILNet

Implementation of ICCV 2017 paper "A Generic Deep Architecture for Single Image Reflection Removal and Image Smoothing"
111 stars 24 forks source link

optim.adam_state #4

Open TengliEd opened 6 years ago

TengliEd commented 6 years ago

In line 230 of training_reflection_ecnn.lua, optim.adam_state was a nil value as I called it. You mean optim.adam?

fqnchina commented 6 years ago

The adam file has been uploaded with notes on how to compile this into the torch framework. Actually, the adam_state function is developed for finetuning the trained framework with existing adam parameters, which is not implemented by Torch before, and doesn't influence the training process. So you can either incorporate our adam file or just use the inherent adam.

Thanks for your question.

yutseng318 commented 6 years ago

Hi, To train the model, I just followed the instructions step-by-step. Put the given adam_state.lua in ./torch/pkg/optim and edit init.lua by adding a new line: require("optim.adam_state") then make. but as I tried to train the ecnn, got the error: luajit: ...er/torch/install/share/lua/5.1/optim/init.lua:35: module 'optim.adam_state.lua' not found: no field package.preload['optim.adam_state.lua'] ...

Did I do something wrong?

TengliEd commented 6 years ago

@yutseng318 You may need to add the new line at the end of the original script.

yutseng318 commented 6 years ago

@TengliEd Tkx