[Question] Patch Based Training

[vmiller987]

The tiling callback only works for specific models.

Is there another way to incorporate patch based training? I have irregular image sizes (2048, 110) that are difficult to feed into models like Ganomaly/Uflow that want a near square image to perform.

Resizing these images ends up with significant resolution loss, or other weird issues. It would be nice if the tiling callback worked for models there.