Higher Learning Rate - Githubissues

jazzsaxmafia / Inpainting

Implementation of "Context Encoders: Feature Learning by Inpainting"

BSD 2-Clause "Simplified" License

183 stars 65 forks source link

Higher Learning Rate #2

Closed pathak22 closed 7 years ago

pathak22 commented 8 years ago

Hi Taeksoo,

I suggest using higher learning for training. In the paper, I used following settings: LR = 0.002 for generator LR = 0.0002 for discriminator I believe it should improve the quality of results a bit. :) Curios to see the updated results.

Thanks, Deepak

jazzsaxmafia commented 8 years ago

Thank you for your advice! I will try again using the parameters you recommended. -Taeksoo