πŸ–ŒοΈ View as RGB images

source

rgb


def rgb(
    x:Tensor, # Tensor to display. [[...], C,H,W] or [[...], H,W,C]
    denorm:Any=None, # Reverse per-channel normalization
    cl:Any=False, # Channel-last
    gutter_px:int=3, # If more than one tensor -> tile with this gutter width
    frame_px:int=1, # If more than one tensor -> tile with this frame width
    scale:int=1, # Scale up. Can't scale down.
    view_width:int=966, # target width of the image
    ax:Optional=None, # Use this Axes
)->RGBProxy:

rgb(image)

rgb(image, scale=2)

two_images = torch.stack([image]*2)
two_images
tensor[2, 3, 196, 196] n=230496 (0.9Mb) x∈[-2.118, 2.640] ΞΌ=-0.388 Οƒ=1.073
in_stats = (    (0.485, 0.456, 0.406),  # Mean
                (0.229, 0.224, 0.225) ) # std
rgb(two_images, denorm=in_stats) # OR: denorm='imagenet'

rgb(two_images, denorm='imagenet')

# This image is in the range [0 .. 1] (+/- eps)
tenchman_01 = image * torch.tensor(in_stats[1])[:, None, None] + torch.tensor(in_stats[0])[:, None, None]

# This one is in the range [-1 .. 1] - also somewhat common
tenchman_minus1_1 = tenchman_01 * 2 - 1
rgb(tenchman_minus1_1, denorm='symmetric') # 'symmetric' does the conversion

rgb(image, denorm='minmax') # 'minmax' scales the full range of the inputs to [0..1] (per-channel)

# Make 8 images with progressively higher brightness and stack them 2x2x2.
eight_images = (torch.stack([image]*8) + torch.linspace(-2, 2, 8)[:,None,None,None])
eight_images = (eight_images
                    .mul(torch.tensor(in_stats[1])[:,None,None])
                    .add(torch.tensor(in_stats[0])[:,None,None])
                    .clamp(0,1)
                    .view(2,2,2,3,196,196)
)
eight_images
tensor[2, 2, 2, 3, 196, 196] n=921984 (3.5Mb) x∈[0., 1.000] ΞΌ=0.382 Οƒ=0.319
rgb(eight_images)

# You can do channel-last too:
rgb(image.permute(1, 2, 0), cl=True)