Starting with square images in ImagesForTraining breaks Colab processing

[mikebind]

I tried to run the colab version of this and preprocessed my images to be square and cropped around the face I was trying to train on, but this seems to have broken the preprocessing done in the Colab. I'm sure the problem comes from hitting some of the width=height conditionals, but I don't understand what each step is doing well enough to know how to fix the problem in a sensible way. Instead I batch resized my square input images to be slightly oblong and restarted. That time all the preprocessing steps went smoothly and I was able to start training. A quick revision by someone who knows what they are doing would be helpful!!

[nicolai256]

if ur images are already square u don't need to do the outpainting preprocessing

[mikebind]

Are any preprocessing steps needed if they're already square and 512x512? Are the mask images used for anything other than resizing? Do they need to be moved to another directory (like the _outpainting directory)?

[mediocreatmybest]

Unsure if this is related but with the squaring of images in the path_to_input_images I'm getting a path is not define error.
for this part "dirs = os.listdir( path )"

NameError: name 'path' is not defined

[mikebind]

That one I knew how to fix. The colab is a bit of a mess and that line just got out of order. The path variable is defined a few lines later in the code, so if you just cut and paste the line giving you the error below that definition it works fine.

[mediocreatmybest]

Awesome, thanks. Python isn't my forte, I did think that was the issue and tried but failed to fix it.
I'll try this again tonight.