Vision: Is there something wrong in normalization

[SunnerLi]

from torchvision import transforms
import torchvision.transforms.functional as F
import torch

# -----------------------------------------------------------------------------------
# Normalize definition in torchvision.transforms.transforms.py
# -----------------------------------------------------------------------------------
class Normalize(object):
    """Normalize an tensor image with mean and standard deviation.
    Given mean: ``(M1,...,Mn)`` and std: ``(S1,..,Sn)`` for ``n`` channels, this transform
    will normalize each channel of the input ``torch.*Tensor`` i.e.
    ``input[channel] = (input[channel] - mean[channel]) / std[channel]``
    Args:
        mean (sequence): Sequence of means for each channel.
        std (sequence): Sequence of standard deviations for each channel.
    """

    def __init__(self, mean, std):
        self.mean = mean
        self.std = std

    def __call__(self, tensor):
        """
        Args:
            tensor (Tensor): Tensor image of size (C, H, W) to be normalized.
        Returns:
            Tensor: Normalized Tensor image.
        """
        return my_normalize(tensor, self.mean, self.std)

    def __repr__(self):
        return self.__class__.__name__ + '(mean={0}, std={1})'.format(self.mean, self.std)

# -----------------------------------------------------------------------------------
# Fundtion definition in torchvision.transforms.functional.py
# -----------------------------------------------------------------------------------
def _is_tensor_image(img):
    return torch.is_tensor(img) and img.ndimension() == 3

def my_normalize(tensor, mean, std):
    """Normalize a tensor image with mean and standard deviation.
    See ``Normalize`` for more details.
    Args:
        tensor (Tensor): Tensor image of size (C, H, W) to be normalized.
        mean (sequence): Sequence of means for each channel.
        std (sequence): Sequence of standard deviations for each channely.
    Returns:
        Tensor: Normalized Tensor image.
    """
    if not _is_tensor_image(tensor):
        raise TypeError('tensor is not a torch image.')
    # TODO: make efficient
    result_tensor = []
    for t, m, s in zip(tensor, mean, std):
        result_tensor.append(t.sub_(m).div_(s))
    return torch.stack(result_tensor, 0)


if __name__ == '__main__':
    # Generate random image which range is (0 ~ 255)
    img = torch.rand(3, 160, 320) * 127.5 + 127.5

    # Define parameter I want to normalize
    my_mean = [127.5, 127.5, 127.5]
    my_std = [127.5, 127.5, 127.5]

    # Adopt official Normalization
    print("max value in tensor before adopt official: ", torch.max(img), img.size())
    norm_op = transforms.Normalize(my_mean, my_std)
    img = norm_op(img)
    print("max value in tensor after  adopt official: ", torch.max(img), img.size())

    # Adopt revision Normalization
    print("max value in tensor before adopt revision: ", torch.max(img), img.size())
    norm_op = Normalize(my_mean, my_std)
    img = norm_op(img)
    print("max value in tensor after  adopt revision: ", torch.max(img), img.size())

Sorry to bother.
Today I try to use normalization function to normalize my data.
However, I cannot get the right result eventually.
As the result, I do the experiment.
The above code shows that I use official transforms.Normalize to deal with the generated data.
The following shows the result:

max value in tensor before adopt official:  254.99990844726562 torch.Size([3, 160, 320])
max value in tensor after  adopt official:  0.9999992847442627 torch.Size([3, 160, 320])
max value in tensor before adopt revision:  0.9999992847442627 torch.Size([3, 160, 320])
max value in tensor after  adopt revision:  -0.9921568632125854 torch.Size([3, 160, 320])

However, after the conduction of the function, the max value of tensor didn't change.
Next, I try to revise as the above edition, and get the right answer finally.
Is there something wrong with my idea?
Or is there some problem in official version?
The version of torchvision I use is 0.2.0

[SunnerLi]

Sorry to bother............
I know what I'm wrong.
The format of tensor should be transform as BCHW,
or the correct result cannot be gained.