[MEncoder-users] Experiences of video encoding

[Corey Hickey]

Mathieu Monnier wrote:
>>  That surprises me: some tests I did last summer, which were a prelude
>>  to these tests, showed that even with lavc, B-frames increase
>>  quality.
> 
> It highly depends on the video content

That is true to an extent; using a vb_strategy can help.

Still, the content is not the actual problem here. For most content, a 
small number of B-frames will improve lavc's encoding quality very 
much--unless you use the vqscale parameter, which disallows lavc from 
using higher quantizers for B-frames and, hence, makes them a waste of 
space.

You'd be better off using lmin and lmax to set the P-frame quantizers 
and let lavc choose I/B-frame quantizers. I'll rewrite and extend the 
table I wrote in an earlier message:

                       I   P   B

XviD fixed_quant=4    4   4   7
x264 qp=4             1   4   6
lavc vqscale=4        4   4   4
lavc lmin=4:lmax=4    3   4   6

If you want, you can further use vi_qfactor, vi_qoffset, vb_qfactor, and 
vb_qoffset to make lavc choose I and B quantizers like XviD or x264. The 
other two codecs have similar parameters.

>>  Thanks, and thanks for your remarks: I will keep them safely for
>>  future reference. But I do not think that this fine-tuning can be
>>  enough for lavc to beat x264, therefore I will keep working on the
>>  latter for now.
> 
> Don't ever expect to beat x264 with a mpeg4p2 codec, at least not if you 
> only compare PSNR & SSIM.

Absolutely. H.264 is a different animal. Still, it is interesting to 
compare them quantitatively.

-Corey