GKnot's vs. internal curve scalling

[molerus]

Hi!

I'm currently compressing "The Terminator" and apart from the 2CD rip I wanted to do 1CD rip to practice. But the results I get are far from my esthetic standards.
My issue is why, while the bright scenes look very OK (too OK), the scene in the disco is horrid. The output file size is 55 % of the first pass. I tried different settings in internal curve scalling and the results were basically similiar. On the other hand some very bitrate-consuming scenes, very detailed, but with low motion look too good.
Now I'm compressing with the external curve scalling, dome by GKnot.
Has anyone compared the results with the internal scalling????
Any suggesttions, observations, hints??
I would be very appreciated :)

P.S.
Is there a way to distinguish complex scenes, where scalling the bitrate down causes noticeable artifacts (for instance flashing light, smoke or flame), from those with rapidly moving objects where the scalling is less unpleasant???

[Vash]

How about offering your settings? What build did you use? What filters? Giving more information would be the best thing to do, just posting "I encoded [...] using [...] scaling and it looks bad" is not enough

[Koepi]

Well, another thing is that you always get horrible results when compressing a MPEG4-2CD once more to MPEG4-1CD.

Believe me, when you try with the DVD sources you'll be amazed about the results, but compressing something highly compressed once more isn't doable in acceptable quality, just as you found out. You're lucky that some scenes look ok.

Regards
Koepi

[iluvatar]

Believe me, when you try with the DVD sources you'll be amazed about the results, but compressing something highly compressed once more isn't doable in acceptable quality, just as you found out. You're lucky that some scenes look ok.

maybe i got it wrong, but i think he's using the dvd as source for the 1cd-rip either. so what does this have to do with recompressing ?

christian

[molerus]

Hi!

@Koepi

You must have misunderstood me. I don't want to recompress 2CD into one but doing the job from DVD source. I just wanted to check, whether it's possible to fit that kind of film into one CD.

@Vash

If you are asking - I encode Terminator using Koepi's build 26-01-2003. The output resolution is 576x312 (sharp bicubic). I use b-frames, 1-100%-200%,lumi masking and chroma ME. I didn't switch the q-pel nor GMC. The output filesize is ~650 MB's (without sound), which is about 55 % of full size after 1-st pass. Used linear curve scalling and 25% boost on keyframes.
Currently I encode without any filters. Since it's fairly dusty film I tried to use the GoldDust filter for AviSynth. Although it improved compressibility (the 1-st pass filesize dropped from 1150 MB to 950 MB), after 2-nd pass the result was pretty much the same (blocks etc) as without any filters. Moreover it took me two days to do both passes with GoldDust.:D

All I wanted is to know whether it is possible to improve curve scalling with the use of GordianKnot, or the internal scalling is better.

[iago]

Moreover it took me two days to do both passes with GoldDust.:D :D

[Koepi]

Please, don't use iframe-boost. It does a good job if you know what you're doing exactly, but for "normal encodes" it does more harm than good.

Use LumaFilter() from mpeg2dec3.dll and UnFilter(-,-) (depending on the source a (+,+) might be better) - i use UnDot() from trbarry with some success as well (inserted before UnFilter()).

Regards
Koepi

[sh0dan]

Some thoughts strike me:

- Try using lanczosresize instead of sharp bicubic - it often produces slightly more compressible result, while being a bit more eyepleasing (subjective opinion).

- Try a temporalsoften(3,3,3,mode=2,scenechange=10) if you are using AviSynth 2.5 beta. It very gently removes slight noise and also give a better compressible image.

- Try adding MergeChroma(Blur(1.0)) - it usually cleans up chroma very nicely.

- Always end with Limiter() (AviSynth 2.5) as the last filter to remove invalid YUV values (helps compression).

Is lumamasking recommended? I've always kept far away from it.

[molerus]

Hi!

Thanks for all your replies.
I'll try the suggested filters.

I currently try to improve the output quality by tweaking the bitrate curve. I was wondering if it wouldn't be appropriate to correct it by moving some bits from scenes with occasionally rising bitrate to those with constant rapid movement, where the results of high compression are more noticeable.
My main concern is whether it would be possible to distinguish movement from moving solid object from 'movement' caused by random brightness changes (flames, strobe-light, smoke).