Le vendredi 08 mai 2020 à 15:16 +0200, Hans Verkuil a écrit :
> > I think this comes directly from the spec. The Rockchip VDEC doesn't
> > seem to use it.
> > Do you think we can drop it from here, and rely on the V4L2 colorspace
> > passed in the format negotiation?
> 
> 
> That would be preferred, yes. Otherwise you would have two places where you can
> define colorspaces, and that's confusing.

This is indeed redundant with the color information in the format. In
VP8/9 there is only 1 header for everything, so it's harder to figure-
out what to filter. While in H.264/HEVC this information is usually set
an an extension header.

Though this do re-open the discussion about SPS in H.264/HEVC. In there
you will find the coded size and the crop window and the sub-sampling.
I don't remember exactly what was the conclusion, but I think it was
kept to allow allow bitstream reconstruction. But it will effectively
overlap directly (or indirectly) with some V4L2 generic API.

regards,
Nicolas