Julien Denize

@edgarriba I rebased on master and applied your suggestions, it should be easier to track the issue

Hi @edgarriba , I've rebased on master the code, from what I found the gradcheck error might be related to how gradcheck basically works. The random for sampling sigma makes...

Sure, but could you tell me what functions you are thinking about, please? I believe it is done for the gaussian part, so are you speaking about other kinds of...

Do you mean for the test that didn't pass or for using sigma as tensor such as in elastic transform? For the former, I checked the list that didn't pass...

Hello, By checking the docstring test that failed, I noticed that for elastic transform it is expected to have sometimes a `requires_grad=True` for `sigma`. I had to update the function...

> LGTM. @ducha-aiki see some failing tests. Maybe this pr needs rebase ? For the failing doctest, there is an issue because sometimes it expects scientific notation sometimes it does...

> you can explore this: https://github.com/kornia/kornia/blob/master/Makefile#L17 > > we have in the makefile several commands useful tor developers Thanks, I've been able to debug the doctest thanks to `make doctest`,...

Hello @edgarriba, thanks for the tests, apparently the problem for the test-lts failures is the Union type hint that I used. Is there any recommendation or convention in Kornia to...

Thanks for merging master in the branch. To fix lts tests, do you have a preference for changing the type hinting of the gaussian blur functions so that Union is...

Hello @edgarriba @ducha-aiki , following your latest remarks, I added new functions that take sigma tensor parameters to remove the `Union[torch.tensor, Tuple[float, float]]` issue. The functions taking float actually convert...