How to load image in default range [0,255] using torchvision?

vision
1

torchvision transforms ToTensor converts the image to [0,1]. Rather than multiplying by 255 afterwards. Can I avoid this normalization in the below code snippet in some way?

transform = transforms.Compose([transforms.ToTensor()])
2

You can do this easily without using ToTensor. All you need is to define your own transform, like this:

class ToTensorWithoutScaling(object):
    """H x W x C -> C x H x W"""
    def __call__(self, picture):
        return torch.ByteTensor(np.array(picture)).permute(2, 0, 1)

transform = transforms.Compose([transforms.ToTensorWithoutScaling()])

Edit: Intermediate conversion to np.array is needed before calling ByteTensor constructor