As part of preprocessing for my image dataset consisting of 16-bit tiff images, I would like to perform a color jitter transformation. I don’t want the network to be too sensitive to brightness in the image. It seems, however, that torchvision.transforms.ColorJitter requires a PIL image, and PIL images must be a uint8 from [0,255]. Is this correct? Should I use skimage or some other libraries to change brightness or is there a way I can get ColorJitter transform to work for 16 bit images in the range [0,65535]. Thank you!