[MPlayer-users] RE: 2 pass with lavc, 2nd pass looks a little worse (only a little)

[Tuukka Toivonen]

On Thu, 14 Aug 2003, D Richard Felker III wrote:
>Hm? This sounds like it's for TV capture, where as Tuukka is ripping a
>DVD. I seriously hope no one rips DVD at 320x240....

The original poster was ripping DVDs (not me), I agree that my options are
finetuned for TV capture and should be modified for DVD.

>> - Use hqdn3d, I use hqdn3d=12:15:15, for good video source decrease the
>>   values. This greatly removes noise (at least when recording from TV)
>Using temporal strength greater than the default (6) will almost
>definitely create ghosts! Take a look and see. This is most noticable

Probably correct for DVD, for TV the noise level is so much higher that it
makes sense to increase the default. Actually, TV hereby is very noisy and
I almost fell from my chair when I saw the result of hqdn3d... it was just
so much cleaner (especially due to temporal filtering).

>may be a worthwhile tradeoff if your input is horribly noisy, but the
>ghosts you get will negatively impact motion estimation too!

Ouch. Why's that? The ghosts are present in every frame, so I wouldn't
expect it to affect ME negatively.

>> vmax_b_frames=2:\
>And using 2 b frames probably both lowers quality AND makes it take
>longer to encode... You should use at most 1 b frame, and imo none...

Actually, I agree. I'll change my script. Two B-frames didn't make it worse
than with no B-frames at all, but it didn't increase the quality
significantly either (I think something like 0.02 dB... I don't remember if
I tried just one B-frame).

Yes, with the options I showed, mencoder was *slow*. You probably can't
make it slower than that.. except with vme=1, but surprisingly that
actually lowered image quality (it definitely should not with rate
distortion optimization cmp=6, so something must be wrong...).

Actually, rate distortion optimization improved quality very little if at
all, I think it should improve it more. Anyone know what's wrong with it?