Do inputs have to be square?

[jbmaxwell]

I wanted to try this with a custom tensor input, but I've hit the asserts in here, which seem to be requiring inputs to be square:

@property
    def resolution(self):
        assert len(self.image_shape) == 3 # CHW
        assert self.image_shape[1] == self.image_shape[2]
        return self.image_shape[1]

Is this actually a hard requirement?

[Zhendong-Wang]

The dataset.py is inherited from EDM Github, which is designed for preprocessed data in squre format. You can use your own dataset file with custom image size, but make sure the height and width are mutliples of 8, since the Unet will downsample and upsample three times.

[jbmaxwell]

Ah, okay, excellent. I'm subclassing your dataset, so I'll just override resolution and remove the square requirement.

Thanks for the speedy reply!