Metrics for epipolar geometry in kornia force the inputs to be batched like below.
if (len(Fm.shape) < 3) or not Fm.shape[-2:] == (3, 3):
raise ValueError(f"Fm must be a (*, 3, 3) tensor. Got {Fm.shape}")
However, I think we can get both batched and unbatched inputs and handle them separately.
Is there any reason behind this logic, e.g. for safety?