[vdr] Re: improving tosvcd 0.9 MPEG quality.

[Steffen Barszus <st_barszus@gmx.de>]

On Tuesday 11 March 2003 21:20, you wrote:
> HI,
>
> Andrew Stevens here... (mpeg2enc author)
>
> I've now got  DTV Card and started playing...
>  is not so bad, at least on TV, but it depends on the material. mpeg2enc
>
> > has some problems with dark movies
>
> Do you have a short snippet of SVCD MPEG-2 that shows the Problem.
> Also, which version of mpeg2enc are you using with tosvcd?

I had the one from mjpegtools 1.6.0 running with tosvcd 0.7

>
> > > I used to create VCDs using an SGI O2 by capturing analog video
> > > material via the O2's capture hardware and by compressing that
> > > into MPEG-1 using SGI's dmconvert utility. The quality of these
> > > VCDs was much, much better than the quality of the SVCDs created
> > > by tosvcd. One would assume the opposite. Should not SVCDs be
> > > better than VCDs?
> >
> > It depends on the compression and how many minutes you try to stuff on
> > one CD.
>
> Here here ;-)
> Actually, if the original feed is a poor quality analog master that has
> then been enocded for DTV and you then transcode that you may get some
> Issues. One Problem is that for SVCD you are scaling down to 480 pixels
> which means you're Macroblock boundaries fall smack in the middle of
> macroblocks in the SVCD image.   This produces big DCT coefficients and
> eats a lot of bits.
>

Would 352x576 be better then ? Sorry I guess i do not understand completly ;) 
(Assuming HQ-DTV/DVB material, the other one would need a filter before the 
encoder)


> > Steven Andrews has begun to work again on mpeg2enc. One thing he is
> > programming on is twopass mode , thats good for better size prediction
> > and better usage of the bitrate. Lets wait which features he tries
> > further to implement. (loog at the mjpegtools-ml-archiv)
>
> Two pass mode won't make a whole heck of a lot of difference for SVCD.  The
> reason is that SVCD is so heavily bit-rate limited it is often running in
> more-or-less as a constant bitrate stream.
>

The whole reason I really would like to have a 2-pass mode is for better size 
prediction. Currently tosvcd restarts the encoding process at the beginning 
if it assumes the size will get to big and takes a higher -q value then. The 
goal is to have 800MB mpeg2 in 2 passes without any restart. Something like 
an average bitrate ( (requested size - audio size)/sec = bitrate or the like) 
As I understood something like an average bitrate can't be reached without 
more then one pass. 

> What probably could make a big difference is a suitably tuned scaler for
> producing the horizontally downscaled image for encoding. What scaler are
> you using in tosvcd?   Depending on the original source material you may
> find using a denoiser may help a lot too.   SVCD's peak bitrate is at the
> limit of what MPEG-2 can do with broadcast material...

As I read the source last time yuvscaler is used. 
mplayer was nice as scaler as I tried it with mencvcd, I didn't tried 
y4mscaler and I didn't looked in tosvcd 0.9 yet. (hadn't much time the past 
weeks)

-- 
Regards
Steffen
____________________
counter.li.org : #296567.
machine: 181800
vdr-box : 87
____________________
Please dont CC me, since if I have replied I'll watch the tread. Both mails 
will be filtered to the ML-folder. Thanks


--
Info:
To unsubscribe send a mail to listar@linuxtv.org with "unsubscribe vdr" as subject.