• torchvision.utils.save_image save a given tensor to an image file

def save_image(tensor, filename, nrow=8, padding=2, 
       normalize=False, range=None, scale_each=False, pad_value=0)

if tensor is on cuda, it will be moved to ``CPU``` automatically.

• torchvision.utils.make_grid generate sprite image (multi-rows, multi-cols)

def make_grid(tensor, nrow=8, padding=2, normalize=False, 
      range=None, scale_each=False, pad_value=0)

tensor: 4D mini-batch tensor [B, C, H, W] or a list of images with the same size
padding: padding between images.
normalize: if True, shift the image to the range (0, 1) by a specific range specified by range
range: tuple (min, max), if this None, minandmax are calculated from the tensor.
scale_each: scale each image separately not use range calculated from all images.