bad result of the training

avalonstrel / GatedConvolution

An reimplement version of inpainting model in Free-Form Image Inpainting with Gated Convolution

Other

115 stars 27 forks source link

bad result of the training #7

Open HolaMan opened 5 years ago

HolaMan commented 5 years ago

Firstly, thanks for the sharing the code

I have trained for a while, reach around 10+ epoch and the result seems not good. I am not sure if I did any wrong on the configuration or others. It seems the result is very weird and can't converge in the long run.

It would be very nice if any hints or suggestion. Thanks in advanced.

JinMengKang commented 5 years ago

I met the same problem. My problem appeared in 3 epoch.Looking forward to the author's answer, the network is hard to train

avalonstrel commented 5 years ago

Sorry, in fact, I also find the problem. Instead of solving the problem by changing the original code, I reimplement this model in pytorch(which is more familiar to me) in https://github.com/avalonstrel/GatedConvolution_pytorch. I provide a pre-trained model on Places2 and some results.