Inconsistent documentation in RandomAffine

In [1]: import torchio as tio
tr = t
In [2]: tr = tio.RandomAffine(scales=1.2)

In [3]: tr.scales
Out[3]: (1.2, 1.2)

But the docs say:

If only one value 𝑑 is provided, \s𝑖∼U(0,𝑑).

@GFabien this was introduced in #244. Do you know what’s going on? Also, note that some wrong latex was introduced in that PR, as you can see in that example: the backslash before s in not necessary (https://torchio.readthedocs.io/transforms/augmentation.html#randomaffine). I’ve fixed it in some other places, but some others might still be around.

[GFabien]

Maybe it’d be good to swing around 0 or 1, depending on the parameter. For example, scale or intensity would go around 1, but degrees or translation would be around 0. For example, if you use scale 1.2, it would be converted to 0.8-1.2, not 0-1.2 (which is definitely not a good idea, I think, as if you get e.g. 0.05 your image is pretty much gone.

This makes sense but I fear this would make things too complicated. It is probably clearer to force the user to pass both min and max values…

[fepegar]

I think it’s fine to use these smart ranges. They use them in torchvision.