@jmnolte Thanks for your contribution. The only downside of implementing the randomness is GridPatch is that this transform is supposed to be non-random while relying on its random counter part RandGridPatch to include randomness. Thus, I would suggest to only modify RandGridPatch for this purpose by overriding its filter_count method, handling situation with sort_fn="random", and delegating the rest to the its parent class.

def filter_count(self, image_np: NdarrayOrTensor, locations: np.ndarray) -> tuple[NdarrayOrTensor, np.ndarray]:
    if self.sort_fn == GridPatchSort.RANDOM:
        pass # do your logic for random sorting here
        return image_np, locations
    return super().filter_count(image_np, locations)

Also it would be great if you can update docstring for RandGridPatchd (the dictionary version of the transform) for this new option ("random") too. Thanks

thanks, it looks good to me -- please help fix the typing error

Thanks for the quick update, LGTM. Could you please help also update the docstring in the RandGridPatchd?

Thanks for addressing the comments. LGTM!