Romash
Romash
Just published the stats. Looks like QOI_COLOR really needs to be reworked.
I implemented QOI_COLOR with RGBA8, RGB8, R4G4B8, R4G8B4, R8G4B4, RG8, RB8, GB8, RA8, GA8, BA8 modes, but I don't really see much difference. Maybe I have chosen the wrong modes?...
At least we still can try to improve the situation by implementing 2 and 4 color diffs in one 24 bit chunk for 12 and 6 bits/pixel compression.
This can move some pixels from chunks DIFF_16 to 12 bits/pixel chunks and from DIFF_8 to 6 bit/pixel chunks.
Although I'd rather try to implement an index diff with a color diff in 1 chunk first.
I implemented a 16-bit chunk with 3-bit opcode, 3-bit index difference from the previous pixel (+-4, no 0) and 433 color difference. It works, but the improvement is really small....
Now I have implemented a 24-bit chunk with 2 color diffs and I think its worth it. Chunk consists of 3-bit opcode, 443 diff and 433 color diff. |Image pack|Original...
I thinking of implementing this encoding inspired by kmar work. 8 bit chunk: 10 + 6 bit index 8 bit chunk: 11 + rgb222 deltas supporting channel expansion 16 bit...
Honestly I didn't understand why increasing the color cache should affect the compression ratio much. This helps mostly for palletized images, but not for regular images where colors are not...
My stats: x | kodak | misc | screenshots | textures | wallpaper | total average -- | -- | -- | -- | -- | -- | -- QOI_INDEX...