ffmpeg colormatrix operations only for 8bpc?

Question

Unless I'm bad at reading, this page states that colormatrix is only good for 8 bpc.

From Colorspace support in FFmpeg:

... colormatrix produces horrible quality for anything > 8bpc (8-bit per component)

Backstory. One of my typical workflows is to convert 16bit tiffs and 10bit DPX frames into 10bit prores. However, I think the input colorspace of the frames is assumed to be bt601 and I would always get encodes that looks just slightly off from my originals. I solved this by adding:

-vf colormatrix=bt601:bt709

And it looks fine visually. The difference between the source and encode I always just assumed was the lossy compression.

So can someone chime in on if the colormatrix=bt601:bt709 operation is 'supposed' to be bad when the source and encode are >10bpc? Or is the info on that page outdated? Or should I be doing it a different way?

Answer 1

Yes, as it says a couple of lines above, colormatrix supports only 8bpc (8-bit per component) pixel formats. So, when a 16-bit input is filtered with colormatrix, it will be converted to 8-bits. And won't get converted back to 16-bit unless manually done or auto-converted due to output encoder limitations. But that's pointless anyway, as the precision has been lost.

Use the colorspace filter.

-vf colorspace=all=bt709:range=tv:format=yuv422p10:iall=bt601:irange=tv

Depending on your use case, format may need to be yuv444p10. Depending in your input, irange may need to be pc.