Hello, I have a 4D tensor of size e.g. (batch_size=10, channels=10, height=6, width=6). I want to resize each (6x6) channel to (30x30). I want to conserve the pixel values as much as possible during resizing, maybe by interpolation or so.
How can I do it for all the channels and images at once instead of one by one with torchvision.transforms?
This might work
from torchvision import transforms
transform = transforms.Compose([
x = torch.ones(10, 10, 6, 6)
y = transform(x)
torch.Size([10, 10, 30, 30])
Thanks @cskarthik7 for reply.
I tried it already, but the problem is that the pixels are a bit dull as you can see in the image which is understandable.
So, any ideas to sharpen the image?
Image resize generally gives an image after resampling. I guess the default resampling method is Nearest Interpolation. You can explore more.
Found rather a better solution through this thread.