RGB tensor to YDbDr

[Dylan_Yung (Dylan)]

YDbDr

Is there a way to convert a RGB tensor to YDbDr? I’m assuming maybe not. This seems like a basic question but I never verified that indexes representing the channels in in the tensor correspond to a color like: 0, 1, 2 = R, G, B, because if so then I should be able to take a tensor of Batch x RGB x Height x Width and use the equation in the wikipedia article.

[InnovArul (Arul)]

Kornia might help you to achieve this.
https://kornia.readthedocs.io/en/latest/color.html#ycbcr

[Dylan_Yung (Dylan)]

is YCbCr the same as YDbDr? Pretty sure this is different.

[InnovArul (Arul)]

Yes, they seem to be different.
You could look at the source code of YCbCr and adapt it to YDbDr.

https://kornia.readthedocs.io/en/latest/_modules/kornia/color/ycbcr.html#rgb_to_ycbcr

@Dylan_Yung

then I should be able to take a tensor of Batch x RGB x Height x Width and use the equation in the wikipedia article.

Either way, to answer your question, Yes. You can take the RGB tensor and use the mapping equation in Wikipedia/elsewhere to convert RGB to YDbDr.

@Dylan_Yung

but I never verified that indexes representing the channels in in the tensor correspond to a color like: 0, 1, 2 = R, G, B

Also, RGB indices depend on the library that’s used to load the images. As I know, Numpy/Scipy libraries load RGB channels in indices 0,1,2 respectively, and OpenCV loads RGB channels in indices 2,1,0 by default.