Reloading Alex LPIPS model every 777 timesteps #4
Description
The LPIPS models seem to get reloaded every 777 timesteps, as if all the globals are then reloaded for some reason. Might be fixed if we place these LPIPS loss model in the Trainer class instead of in the globals. Also strange that it seems to be loaded 4 times everytime, does this have to with the amount of workers?