Size mismatch issue when using --image_size != 128

[lucidrains]

Size mismatch issue when using --image_size != 128

[lucidrains]

I'll try to make this error more informative, but basically the trained model can only generate images on the same exact setting that you trained it on. This includes not only image size, but also --network-capacity. I'll think about how I can make this more seamless and intuitive moving forward

[zax9]

@lucidrains I found this when searching previous issues before opening a new issue for a similar experience I was having. Is it possible to detect the image size and network capacity from the .pt file and use those values instead of the defaults, if they aren't specified? Maybe with an accompanying message e.g. "Network size detected as 32, overriding default value of 16" and "Image size detected as 512, overriding default value of 128"? (Just trying to think of ways to make it clear what the default is and that it isn't being used--if, of course, it can be detected from the .pt file)