search

SHI-Labs / StyleNAT

New flexible and efficient image generation framework that sets new SOTA on FFHQ-256 with FID 2.05, 2022
MIT License
99 stars 11 forks source link

Loading the trained model #14

Closed EskildAndersen closed 6 months ago

EskildAndersen commented 7 months ago

Where in the code are one supposed to load a pre-trained model for evaluation on let's say FFHQ-256?

stevenwalton commented 7 months ago

What issue are you running into? Did you check the configs?. The inference config has a minimal example of how to run inference, which includes loading a checkpoint). I also created meta_conf which has all possible config options in an annotated form. Note that this won't work as its documentation, not an actual config.

If you want to evaluate, you should set type to evaluate. main file will load the generator's checkpoint (if restart.ckpt is defined) and then decide which method to run. Training, inference, or evaluate. You can find the evaluation file here.

After reading the documentation (make sure you also read the README file), if you're still having problems let me know. Please provide the specific error and I'll do my best to help.

EskildAndersen commented 6 months ago

Thanks for the help, we made it work!

stevenwalton commented 6 months ago

Just to confirm, was this the solution? Or was it something else?