search

ZFTurbo / volumentations

Library for 3D augmentations
MIT License
232 stars 37 forks source link

data type in augmentation #32

Open jizhang02 opened 7 months ago

jizhang02 commented 7 months ago

Hello,

I'm curious whether the data type is limited to uint8, ranging from 0-255. In certain situations, such as medical imaging, the data type might be float32. Converting it to uint8 could result in a loss of information.

When I use import torchvision.transforms, it also only support uint8.

ZFTurbo commented 7 months ago

I think it's because transforms came from 2D images, where 0-255 range is typically used for RGB images. May be volumentations can be updated to work with float32.