Quoting Martin Storsjö (2015-04-30 19:33:43)
> On Thu, 30 Apr 2015, Anton Khirnov wrote:
> 
> > Quoting Martin Storsjö (2015-04-29 17:01:53)
> >> ---
> >> The input option handling probably isn't quite right. This probably
> >> also should have some documentation somewhere?
> >> ---
> >
> > I am mostly fine with this patch as it is as an initial implementation
> > (+some docs) and I agree that it should be enabled by default. It should
> > also be a per-stream option I think.
> 
> Can you give hints on how to change it into a per-stream option? This was 
> the best proof-of-concept I managed to do while trying (for a few 
> minutes) to make sense of the avconv/cmdutils options declarations.

Pretty much duplicate e.g. ts_scale -- SpecifierOpt* in OptionsContext
and MATCH_PER_STREAM_OPT() in add_input_streams().

> 
> > Also, we currently send this side data to the decoder, where it
> > typically gets turned into frame side data and survives to filters, or
> > possibly even the muxer. That should not happen if we already applied
> > the rotation.
> 
> Indeed, we should probably clear this piece of side data if we consume it 
> in this way. What's the best place in avconv to do that?

The easiest way for now is probably explicitly excluding it in
process_input().

> 
> > Another thing we should keep in mind for the future is that the rotation
> > side data can also be present on the codec level or the packet level --
> > the former is suported by h264 and hevc (but probably not used in the
> > wild much or at all),
> 
> You mean it is possible to signal this via some element within SPS/PPS/SEI 
> somewhere?

Yes, there's a SEI type for this.

-- 
Anton Khirnov
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to