Deem the lower part of an Image in PyTorch

vision
1

I wish to deem a part of an image. Not completely black but just creating shadows in the lower regions.

2

You just have to bring the pixel values of the tensor that represents the image closer to zero. Something like this:

import torchvision
img = Image.open("cat.jpg").resize((224, 224))
img_tensor = torchvision.transforms.ToTensor()(img)
img_tensor[:, 180:, :] *= 0.75  # reduce the brightness of everything below pixel 180 by 25%
img_shaded = torchvision.transforms.ToPILImage()(img_tensor)

original:
image

after shading:
image

3

Wow, thanks a lot. This is exactly what I wanted.

4

What if I don’t always intend to use a straight line. E.g a wavy line or some sort of randomized line going from one point to another endpoint but not a straight line?

5

In that case you need to replace this with some fancier, custom logic:

img_tensor[:, 180:, :] *= 0.75  # reduce the brightness of everything below pixel 180 by 25%

It will come down to finding clever ways to parameterize the behavior you’d like without too much work. For example for a wavy line you might do:

img = Image.open("cat.jpg").resize((224, 224))
img_tensor = torchvision.transforms.ToTensor()(img)

period = 80
amplitude = 20

for n in range(224):
    i = amplitude * math.sin(n / (period / (2 * math.pi)))
    img_tensor[:, (180 + int(round(i))):, n] *= 0.75
img_shaded = torchvision.transforms.ToPILImage()(img_tensor)

image

1 Like
6

Thanks so very much Andrei. You’re such a gem. Everything is perfect the way I want it now. Thanks for your reply.

1 Like