Brats transform float32 versus int8

[mioDio]

Hi,

I'm experimenting with your tutorials and the Brats Jupyter notebook contains the data transform called "ConvertToMultiChannelBasedOnBratsClassesd". I'm wondering why the transform is using np.float32
to convert the masks to the labels. Is it correct that each channel will only have 1s and thus it could just be a np.unit8?
Wouldn't that save considerable memory?

Happy to be corrected
Cheers

[Nic-Ma]

Hi @mioDio ,

Thanks for raising this ticket, @wyli already submitted PR #1671 to update it.
Need @IsaacYangSLA to help verify locally with Brats dataset.

Thanks.