Nicolas Hug comments

Results 570 comments of


                                            Nicolas Hug

Setting a 2D tensor to `RandomAffine()` after instantiation got other error message against error message

Thanks for the report - if there are ways to keep the code simple and improve the error messages for cases that make sense (which, I'm not entirely sure is...

For `kernel_size` parameter, the doc of `GaussianBlur()` doesn't explain which is width or height

It will always be H and W as in pytorch, but feel free to submit a PR to make this clearer

`alpha` argument of `ElasticTransform()` should completely avoid negative values, giving error and the doc should have the explanation.

I guess we can document this, but maybe raising would be disruptive.

`sigma` argument of `ElasticTransform()` should completely avoid negative values, giving error and the doc should have the explanation.

Happy to consider a PR for the docs. Not sure we need to raise at this point, considering there might be existing usage which we would be breaking

the doc of `alpha` and `sigma` argument of `ElasticTransform()` should have the explanation about if it's a single value

Agreed this should be documented, happy to consider a PR. Thanks @hyperkai

Setting a single negative value or a couple of negative values to `sigma` argument of `ElasticTransform()` doesn't get error against the error message

happy to consider a PR @hyperkai .

The explanation for `all_ops` parameter of `AugMix()` is not clear enough.

Thanks for the report @hyperkai , feel free to submit a PR to clarify the docs.

`NEAREST_EXACT` and `BICUBIC` work against the doc of `AutoAugment()`

Feel free to submit a PR to clarify the docs @hyperkai .

Setting a list of no transforms to `transforms` argument of `RandomApply()` gets the indirect error message

Thanks for the report, feel free to submit a PR @hyperkai

`BICUBIC` works against the doc of `ElasticTransform()`

Feel free to submit a PR to clarify the docs @hyperkai .