[MPlayer-users] Re: file size falls short of 700MB for multi pass

[Jonathan Rogers]

Jason Pepas wrote:
> hmm, which bitrate is it that you are increasing?  are you increasing the 
> bitrate for both video passes, or just the final video pass?
> 
> I just tried the following:
> 
> - audio pass
> - prints out suggested bitrates
> - video pass 1, using higher-than-suggested-bitrate
> - video pass 2, using same bitrate as video pass 1
> 
> and it still came out the exact same size as using the suggested bitrate.

I've usually just redone the second pass, increasing the requested 
bitrate. Come to think of it, if the rate control is delusional, it's 
not surprising that your method doesn't help. Maybe you have to trick it 
by using two different bitrates.

The other thing I haven't really been clear on is whether to use a 
constant quantizer. The mplayer manpage says it's a good idea for the 
first pass, but sometimes it seems to throw things off. For a while, I 
was specifying vqscale=2 for the first pass and vbitrate for the second, 
thinking that the bitrate for the first pass wasn't important. Then, I 
realized that in the absence of vbitrate=x, it just assumes 800.

So, for the first pass, there are actually four permutations:

vqscale=2, no vbitrate (defaults to 800)
vqscale=2, vbitrate=desired bitrate
no vqscale, no vbitrate (defaults to 800)
no vqscale, vbitrate=desired bitrate

My first guess would be that both passes should have identical options, 
except for vpass. However, the manpage recommends vqscale. The actual 
bitrate always goes way over the requested (or default) one when 
specifying vqscale, which implies that the requested bitrate may not 
matter much. Or, maybe it matters a lot. As I try more things, I just 
become more confused.

I was just reading the manpage about "vrc_eq" and noticed that the 
current default equation is simpler than the "old" one. The fact that 
the current default doesn't use any averages means that it's not taking 
any history into account. This implies that the second pass is not going 
to do any better hitting the target than the first. Of course, you 
mentioned that increasing the bitrate for both first and second passes 
results in no increase in the actual average bitrate, which implies that 
the second pass is paying attention to something. I guess we should 
really be asking these questions on the ffmpeg list.

Jonathan Rogers