[MPlayer-users] 2 pass encoding / encode sound twice?

[Ville Saari]

On Tue, Apr 20, 2004 at 12:18:14PM -0400, D Richard Felker III wrote:

> Forget about B frames entirely. It's a good idea never to use them
> anyway.

To my experience that's certainly not true or otherwise there has been
something grossly wrong in the encoding options I have used in my
comparisons I also have experience of using mbd=1 vs. mbd=2 that
contradicts the general opinion of the people on this list.

> Even with just I/P you have a huge problem. Suppose you have a
> sequence of frames like this:
> 
> .....|..
> 
> where the dots are all similar frames and the pipe is a one-frame
> flash of something different. The number of bits to encode the . after
> the | changes RADICALLY if the | frame is dropped!

Yes, but the number of bits used for the second . after the | or all the
rest of the stream would be unaffected. The effect would be local. In this
example only the first . after | would be underquantized, which is quite
insignificant epecially when the viewer is preoccupied by the outrage
caused by the complete loss of that explosion special effect!

> Keep in mind that although these effects are small, so is everything
> in video encoding. Having the wrong pass1 info could make a much
> larger difference to final PSNR than options like trell, mv0, 4mv,

Only if it is wrong for a significant number of frames. If the second
pass is able to resynchronize, then the few affected frames can't do
very much to the size or the average PSNR or the whole file and any
visible defects caused by misquantization are masked by much bigger
annoyance caused by the lost or duplicated frames.

-- 
 Ville