Open bycloudai opened 4 years ago
Have same error. Please help!
Change this parameter to 48: https://github.com/Lotayou/Face-Renovation/blob/b5f5e9e86ec8ebdebcf64d22d806085fead80b8b/options/config_hifacegan.py#L59 It controls the input channel, 48 means 48x16=768 channels. https://github.com/Lotayou/Face-Renovation/blob/b5f5e9e86ec8ebdebcf64d22d806085fead80b8b/models/networks/generator.py#L195-L206
The training will work in this case, but I don't know if the result will be good... The code seems too messy...
@kwea123 Thanks for mentioning. The code has now been reformatted and should work for both 48 and 64. https://github.com/Lotayou/Face-Renovation/blob/2e55e1563d52c6cfb2bbd98c915ae95f566f416d/models/networks/generator.py#L195-L206
Most experiments reported in the paper are trained with ngf=48
that fits on a 12GB Titan X card, and the result is already good enough to beat SOTA. The final code release is tested on a company server with 16GB P100 card, allowing training with ngf=64
. I haven't benchmarked the performance between 48 and 64 though.
Hello again, I got this training error when running "train.py", how can I solve this?
I've done exactly what you described for degrade.py, input 512x512 image and produce a paired image which is 512x1024.
Here's the training config
thank you!