MPEG quant "noise"..intended or side-effect?

[cjv]

Hi,

I've recently been doing some experimenting encoding my fav. R1 movies (2cd, w/AC3)..so approx 1200 bitrate for most..using XviD-27072002-1..but I've come up with a few questions.

I use avisynth bicubic resize 0.6 with MPEG quant for both passes, and the quality/detail is OUTSTANDING..but I've never been fond of the additional "pixel noise" that appears from using MPEG quant.

If I do both passes with h.263, sure enough the noise is gone, and it does look a little smoother, but am I really losing detail?..or is it just appearing smoother because the "artificial" noise added by the MPEG quant is not present. Basically, is the "noise" added by the MPEG quant intentional, or is it just a side effect? And which of h.263 or MPEG is more technologically advanced..or are they just different approaches?

If it is indeed intentional, then I suppose using h.263 and adding a user-controllable amount of noise via ffdshow would achieve the same effect and give the best of both worlds.

Thanks,

cjv

PS I know I could use modulated, but since most of the quants are < 4, I would be using almost entirely MPEG, and thats back to the noise again.

[tanksimpson]

The pixel noise is a side effect. Some people like it that way because it looks more like the original MPEG-2, I guess. IMHO, I don't think the MPEG quantizer offers any advantage unless you are encoding a lot of frames with a quant of "1", otherwise I think H263 gives you better quality at a given bitrate. I would only use the MPEG quantizer if I was encoding at ~3000kbps and up - but I'm curious to hear what the experts say. If you want to test my theory, then encode the same clip twice at a fixed quantizer of "1", first with MPEG, then with H263 - you'll see that the MPEG clip is much smaller in file size. Then you can do the same test with a fixed quant of "2", and I bet both clips will be about the same size.

[Bulletproof]

That's why I said before that deciding on H263 or MPEG should not be through quantizers. I said before that maybe there was a way by looking at the luminance of a picture and figuring out it's complexity. And if it reaches a certain threshold, then decide which is better to use, H263 or MPEG. Like scenes with high complexity would get MPEG and scenes with low complexity H263 (Better yet, make that an option). Cause as of right now, if a scene is low complexity it will probably be given a 2X quantizer which will use MPEG. I don't know the difficulty of adding this, but there is already luminance examining code from the luminance masking code (Pure guessing here).

Another thing I was thinking about is being able to actually tweak the luminance masking, giving it "Strength" and "Sensitivity" settings.

And also, I was wondering, currently the default maximum bitrate set in XviD is 10000, Is this really necessary? I'm assuming it never actually gets this high but wouldn't it be better to just force it to like 2000 because it's hard to actually tell a difference above that (depending on the resolution).

[OUTPinged_]

That's why I said before that deciding on H263 or MPEG should not be through quantizers.

*nods* The thing is, no one did really state which frames will better look with MPEG and which with h263 quantizers (considering a possible quantizer value reduction because of a tad beter compressibility).


And also, I was wondering, currently the default maximum bitrate set in XviD is 10000, Is this really necessary?

That pretty much depends on the encode's bitrate and clip resolution. If you'd try to encode a HDTV stuff, you'd notice that bitrate can jump over 10k pretty often.

The CC of xvid is pretty good so that you didn't have to tweak that value for a typical dvd rip.

[Koepi]

Well, well, funny to read my experts discuss things which are out of their scope.

Who are you to tell us that modulated quant type shouldn't be based on quants?

What do you know?

I sure won't reply to this thread to tell you the technical differences, as the answer was discussed to death in the past, sorry if you can't do a successful search.

[Bulletproof]

Koepi, you think too highly of yourself and you are most arrogant.

"Who are you to tell us"

"What do you know?"

I don't think it is right that you take the opinion for others that contribute to the XviD team. YOU may not want to answer the question, but that doesn't mean others can't.

[Koepi]

Well, you come up with strange stuff very often for my taste bulletproof, so I'm sometimes a little allergic against such proposals.

I herewith apologize.

Still, if you'd search for "modulated quantizer" "threshold" etc. you'd find the thread where we discussed this. There you'll find all the answers which are in question here.

Regards,
Koepi