[Fix]: fix (Cached) MixUp padding

[plutoyuxie]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

[CLAassistant]

All committers have signed the CLA.

[ZwwWayne]

Hi @plutoyuxie ,
Thanks for your PR. Could you explain more.about the motivation of this PR, e.g., why do you need to change the padded part should be ones instead of zeros？

[plutoyuxie]

Hi @plutoyuxie , Thanks for your PR. Could you explain more.about the motivation of this PR, e.g., why do you need to change the padded part should be ones instead of zeros？

Dear @ZwwWayne，below is my motivation of this PR:

When I use browse_dataset.py to visual check CachedMixUp augmentation in my own project, some kind of artificial grey strips appeared.
The padded area is not filled with the given pad_val properly.
It is not clear in coco dataset, but very obvious when a x-ray dataset was used in my project, while the image background is white and pad_val=255.0 is needed in my case.