Re: [rtcweb] VP8 / H.264 conversion without transcoding

The primary advantage you get from this technique is the ability to skip or dramatically reduce motion estimation on the encoder.  The algorithms are similar at a high level but different enough at a low level to make a light weight transcoder impossible

----- Original Message -----
From: Monty Montgomery [mailto:xiphmont@gmail.com]
Sent: Thursday, November 21, 2013 02:00 PM
To: Bossiel <bossiel@yahoo.fr>
Cc: rtcweb@ietf.org <rtcweb@ietf.org>
Subject: Re: [rtcweb] VP8 / H.264 conversion without transcoding

On Thu, Nov 21, 2013 at 2:49 PM, Bossiel <bossiel@yahoo.fr> wrote:
> Entropy decoding (CABAC/CAVLC) must be done to get the macroblock
> structure(partirions, motion vectors...) and residual (AC/DC coeffs). The
> only part you're skiping is the inverse transform, scaling and motion
> compensation.

You wouldn't even be skipping the inverse transform.  VP8 and h.264
don't use the same DCT, so you'd have to do an inverse to track
transform error propagation, which won't be the same in the two
codecs.  Unless, of course, something else rather clever is going
on...

Monty
_______________________________________________
rtcweb mailing list
rtcweb@ietf.org
https://www.ietf.org/mailman/listinfo/rtcweb