3 ms·

This is fine for real world videos crammed into gifs. But please think of animated pixel art, which is antithetical to many the assumptions that DCT-block based

This is fine for real world videos crammed into gifs. But please think of animated pixel art, which is antithetical to many the assumptions that DCT-block based codecs make. Doubly so when you consider that most encoders default to 4:2:0 and ignore transparency. APNG or similar image formats are better for this purpose

bscphil 8y ago
>Doubly so when you consider that most encoders default to 4:2:0 I agree, but I want to add that this has another side effect you don't mention: it makes high quality images impossible. In 2019, it would nice to have something better than cjpeg -q 98 for near lossless (or completely lossless images). I've spent many hours tinkering with the new inter frame based image codecs, and I haven't been able to achieve this because despite their advertised support for better standards, the decoders seem to have only implemented 8 bit 4:2:0 subsampled sRGB images. I guess I'll go on storing my photography in 150 MB TIFF files...
jimbo1qaz 8y ago
Good: - AV1 has a palette predictor designed to improve on pixel art by storing individual pixels as indexes into a palette. - Chroma from luma helps preserve exact colors in pixel art, assuming color is lonely related to brightness (won't work as well for color-only changes, possibly not when brightness and color vary separately, or multiple colors and brightness are mixed) Bad: - AV1 takes seconds to minutes per frame. I let my laptop chug for half to several hours, running Manjaro AUR with bleeding-edge ffmpeg-git, encoding multicore tiled AV1, and it managed to encode under 3 seconds of video. (Did it link to an older libaom-av1 or is it bundled statically with ffmpeg?) - When chroma subsampling and CfL (chroma from luma) mix, the brightness is subsampled to create a color prediction, instead of being used to add extra color resolution. - Youtube's sample AV1 encodes have 4:2:0 subsampling (aka color channels have halved width and height resolution). Rav1e may be faster, but apparently disabling chroma subsampling breaks CfL and inter-frames.

bscphil 8y ago

>Doubly so when you consider that most encoders default to 4:2:0 I agree, but I want to add that this has another side effect you don't mention: it makes high quality images impossible. In 2019, it would nice to have something better than cjpeg -q 98 for near lossless (or completely lossless images). I've spent many hours tinkering with the new inter frame based image codecs, and I haven't been able to achieve this because despite their advertised support for better standards, the decoders seem to have only implemented 8 bit 4:2:0 subsampled sRGB images. I guess I'll go on storing my photography in 150 MB TIFF files...

jimbo1qaz 8y ago

Good: - AV1 has a palette predictor designed to improve on pixel art by storing individual pixels as indexes into a palette. - Chroma from luma helps preserve exact colors in pixel art, assuming color is lonely related to brightness (won't work as well for color-only changes, possibly not when brightness and color vary separately, or multiple colors and brightness are mixed) Bad: - AV1 takes seconds to minutes per frame. I let my laptop chug for half to several hours, running Manjaro AUR with bleeding-edge ffmpeg-git, encoding multicore tiled AV1, and it managed to encode under 3 seconds of video. (Did it link to an older libaom-av1 or is it bundled statically with ffmpeg?) - When chroma subsampling and CfL (chroma from luma) mix, the brightness is subsampled to create a color prediction, instead of being used to add extra color resolution. - Youtube's sample AV1 encodes have 4:2:0 subsampling (aka color channels have halved width and height resolution). Rav1e may be faster, but apparently disabling chroma subsampling breaks CfL and inter-frames.