-
-
Notifications
You must be signed in to change notification settings - Fork 259
Description
Describe your use-case.
All of the images in my dataset are already the right size, and both dimensions of all images are divisible by 64.
I don't want the software to resize them, I just want to train on the sizes they already are.
But when I add a concept with images in 3:4 aspect ratio they are put into a bucket with an aspect ratio of 5:4 according to the "aspect_buckets" section of the config json (even if the concept has resolution override with the correct resolution). Or if I disable aspect ratio bucketing it won't run at all. ( "invalid literal for int() with base 10" error )
What would you like to see as a solution?
Is it possible to have a switch to completely skip all of the aspect ratio calculations and resizing and just make the buckets based on the existing (already correct) image size ?
Have you considered alternatives? List them here.
No response