[x264-devel] Google detailed test results for VP8 to H.264 comparisions

[Jason Garrett-Glaser]

On Wed, Mar 13, 2013 at 10:51 AM, Sergio Garcia Murillo
<sergio.garcia.murillo at gmail.com> wrote:
> El 13/03/2013 18:20, Jason Garrett-Glaser escribió:
>
>> By and large the comparison looks mostly just incompetent; beyond the
>> obvious "not setting both encoders to tune for the same metric" and
>> "using baseline instead of high even though high is what everyone
>> would actually use in practice", they didn't even set appropriate VBV
>> parameters on x264 despite setting some on VP8.
>>
>
> What would be the best encoding parameters fro x264 in these scenario?

1.  Set the slowest preset you're willing to tolerate.
2.  Set --tune to psnr or ssim (based on the one you're testing).
3.  Since they're doing CBR streaming for WebRTC, they need to set
appropriate ratecontrol settings.  Some examples:
1 second of latency: --vbv-maxrate $b --vbv-bufsize $b
0.25 seconds of latency: --vbv-maxrate $b --vbv-bufsize $b/4
low latency streaming with capped frame size: --vbv-maxrate $b
--vbv-bufsize $b/fps

In addition, if they care about latency, they'd want to make sure both
encoders are set with appropriate low-latency settings.  They'd also
need to verify how well both encoders are abiding by these
constraints.

Jason