Implement RGB to YCgCo-R #1572

Activity

1. farindk

   added a commit that references this issue on Aug 18, 2025

   implement RGB to YCgCo (Y/C equal bit depths) #1572

   3d555a2
2. novomesk commented on Dec 24, 2025

   

   novomesk

   on Dec 24, 2025

   More actions

   I have AVIF with matrix_coefficients=16

   http://188.121.162.14/zoo.avif

   heif-dec gives following output:

   

   output of avifdec:

   
3. farindk

   added a commit that references this issue on Dec 26, 2025

   add (incomplete) implementation of YCgCo-R color transformation (#1572)

   86f09c7
4. farindk commented on Dec 26, 2025

   

   farindk

   on Dec 26, 2025

   MemberAuthor

   More actions

   The above is a first, quick implementation of YCgCo-Re (matrix_coefficients=16). It is probably still incomplete, but works for the example image. A better implementation is still future work.
5. farindk commented on Dec 26, 2025

   

   farindk

   on Dec 26, 2025

   MemberAuthor

   More actions

   Note: the example image's pixi box says that the image has 10 bits per pixel:

   | | | Box: pixi ----- (Pixel Information)
   | | | size: 16   (header size: 12)
   | | | bits_per_channel: 10,10,10
   

   This is the internal YCgCo-Re space. The desired output RGB depth is 8 bits per pixel. How can we communicate this cleanly through the API?
6. farindk

   added a commit that references this issue on Dec 26, 2025

   limit color conversion function 14 bpp (#1572)

   e9b3f68
7. novomesk commented on Dec 26, 2025

   

   novomesk

   on Dec 26, 2025

   More actions

   I understand it this way:

   The -RE indicate that "even" (2) bits were added to intended bit depth,
   so 10-2=8 is desired RGB depth for the output.
8. farindk commented on Dec 26, 2025

   

   farindk

   on Dec 26, 2025

   MemberAuthor

   More actions

   I understand it this way:

   The -RE indicate that "even" (2) bits were added to intended bit depth, so 10-2=8 is desired RGB depth for the output.

   Yes, correct. My problem is that the color-conversion function where I added it, must have the same output bpp as the input. And the whole conversion pipeline does not expect that it is desired to have less bits in the output than in the input. This needs more work. But at least, we get a usable picture now.