[FFmpeg-devel] [RFC] QP table API

[Michael Niedermayer]

On Tue, Mar 10, 2020 at 06:45:56PM +0100, Anton Khirnov wrote:
> Hi,
> sending a proposed new API for exporting QP tables (and possibly other
> things in the future). It is based on a patch from Juan De León
> (cf. <20190819233655.171637-1-juandl at google.com>).
> 
> The set is intended for dicussion, not merging at this point. It does
> not adapt the filters and thus breaks FATE.
> 
> Some points to discuss:

> - should the per-block structure store the final per-block quantizer or
>   the delta from the per-frame value (each can be easily computed from
>   the other, so it's more of a convenience question)? deltas are used in
>   the current set as suggested by Lynne

If deltas are used then it should be documented how they are applied to
find the final value.
some codecs might use something like (global + delta) % M in the bitstream, which
seems not ideal for generically interpretable data


> - do we want per-plane quantizers in each block? The original patch
>   didn't have them. Do we want separate AC/DC quantizers? Anything else?

I suspect whatever we put in now, one day an important codec will require
an additional value per block. Thus i think its important that its
extendible. I think the patch was extendible so thats probably already
fine, iam just mentioning it as its important IMHO



> - what to do with (M)JPEG? AFAIU it doesn't have a quantizer scale like
>   later codecs, just full quantization matrices. Our decoder currently
>   exports some number calculated from those in a non-obvious way. Is
>   that useful or should that be dropped?

postprocessing improves (low bitrate) jpeg. So something that provides
information about the quatization step size would be usefull.
In which way that is done isnt important. Exporting the quantization
tables themselfs via side data would be the obvious option

thx
[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Those who are best at talking, realize last or never when they are wrong.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20200311/4d798137/attachment.sig>