 Did I denormalize wrongly?

With reference to discussion here over how to denormalize an image, I reckon I can code up my own just by using the formula. But the images shown are all fuzzy and incorrect.

I am just using simple math formula to revert the normalization to show, did i make any mistake?

def unnormalize(
normalized_img, mean, std, max_pixel_value=255.0
) -> torch.Tensor:
"""TODO: Use https://discuss.pytorch.org/t/simple-way-to-inverse-transform-normalization/4821/7 code and make it a class to include both Normalize and Unnormalize Method.

Formula:
Normalize: img = (img - mean * max_pixel_value) / (std * max_pixel_value)
Unnormalize: img = img * std * max_pixel_value + mean * max_pixel_value

Args:
normalized_img ([type]): [description]
mean ([type]): [description]
std ([type]): [description]
max_pixel_value (float, optional): [description]. Defaults to 255.0.

Returns:
torch.Tensor: [description]
"""
# normalized_img = (unnormalized_img - mean * max_pixel_value) / (std * max_pixel_value)
# unnormalized_img = normalized_img * (std * max_pixel_values) + mean * max_pixel_values

unnormalized = torch.zeros(normalized_img.size(), dtype=torch.float64)
unnormalized[0, :, :] = (
normalized_img[0, :, :] * (std * max_pixel_value)
+ mean * max_pixel_value
)
unnormalized[1, :, :] = (
normalized_img[1, :, :] * (std * max_pixel_value)
+ mean * max_pixel_value
)
unnormalized[2, :, :] = (
normalized_img[2, :, :] * (std * max_pixel_value)
+ mean * max_pixel_value
)

return unnormalized

# TODO: Consider adding plot size as in notebook it is too small.
def show_image(
nrows: int = 3,
ncols: int = 4,
mean: List[float] = [0.485, 0.456, 0.406],
std: List[float] = [0.229, 0.224, 0.225],
one_channel: bool = False,
):
"""Plot a grid of image from Dataloader.

Args:
train_dataset (torch.utils.data.Dataset): [description]
nrows (int, optional): [description]. Defaults to 3.
ncols (int, optional): [description]. Defaults to 4.
mean (List[float], optional): [description]. Defaults to None.
std (List[float], optional): [description]. Defaults to None.
"""

one_batch_images, one_batch_targets = (
dataiter.next()["X"],
dataiter.next()["y"],
)

# TODO: FIX UNNORMALIZE not showing properly.
one_batch_images = [
unnormalize(image, mean, std, max_pixel_value=255.0)
for image in one_batch_images
]

# create grid of images
image_grid = torchvision.utils.make_grid(one_batch_images, normalize=False)

if one_channel:
pass

image_grid = image_grid.numpy()
if one_channel:
plt.imshow(image_grid, cmap="Greys")
else:
plt.imshow(np.transpose(image_grid, (1, 2, 0)))
plt.show()

return image_grid