low bitrate tweaks?

[Blue_MiSfit]

Hey all,

So I'm a master of MPEG-4, but don't know a whole lot about squeezing max quality out of an MPEG-2 encoder like CCE.

I have some PAL stuff that has to fit on a DVD5 - about 2 hours and 5 minutes, which with a 192kbit AC3 works out to an average bitrate of ~4600kbit. This is a bit on the low side, and I have noticed in an unfiltered (aside from NTSC->PAL conversion) encoder that a lot of detail is lost when compared to an NTSC encode at ~8000kbit. Now, obviously there's a huge difference in bitrate, and I don't expect them to look the same, but I would like to squeeze as much quality out of it as I can.

I added fft3dfilter(sigma=2, plane=4) to the script, and removegrain(mode=5), and that gave me a handy 15-20% compressibility boost (at least with an xvid comptest). I am wondering how to tweak CCE for lower bitrates, and hope that you gurus out there can offer me some insight.

The material is high motion with a mostly fixed background, it's a long exercise video of people on an exercise bicycle - from 3 angles with the occasional flying shot on a jib, so for the most part I would call it low motion.

Perhaps a good CQM?

~MiSfit

[G_M_C]

I'm no guru at CCE, but i do use DVD-RB with CCE sometimes and use AviSynth quite often these days.

What i would do is changing the FFT-filter call into sigma=0,sigma2=1,sigma3=2,sigma4=2 .... <rest of the options>.... . That would leave the edges/sharp transitions unfiltered while filtering low-freq (for instance the blue sky) with sigma2. That might give the idea of having more details. The removegrain ... dont know if that is nesessary, only with a grainy source offcourse; Tweaaing the FFT might give similar results, without loosing more details because of the multiple filtering.

Secondly would use one of the custom matrices that come with DVD-RB's matrix-editor. There are several included with that program, that are suited for bitrates > 4000, but allways do a multiple pass encoding. I think that would help much when preserving detail :)

PS: You might switch to FFT3dGPU if your system can run it. It's the GPU version of FFT3Dfilter and has the same possibillities, but it is much faster when used with more modern graphics boards.Look at it here: http://forum.doom9.org/showthread.php?t=89941

[Blue_MiSfit]

Indeed. <3 fft3dgpu, but I'm running under Parallels on Mac OS X, which means that CPU, RAM, and HD access is native, but no direct access to the 7300gt in my PCIe slot :)

I've been having a lot of issues with CCE hanging, both in native Windows XP on Boot Camp and in Parallels. I'm using QuickTime files as a source, so that could be the problem. Anyone have experience with this kind of thing?

[Chainmax]

I don't think there are many ways to squeeze IQ for low bitrate encodes unless you go the "Getting the best out of CCE" route. Here's what I've picked up so far:

- Custom Matrices:
AVAMAT6: Full-D1 with >3500kbps

AVAMAT7: Full-D1 with 1800<bitrate<3500

QLB: Half-D1 with ~1360kbps

- Set "Flat part priority" to 30

- Progressive content should use zigzag scanning

- Set "DC precition" to 10

- Set GOP Sequence's M to 3 and N/M to 4

- Use 3 passes to squeeze the most out of the bitrate curve and set minimum bitrate to 0

By the way, 4600kbps is plenty for Full-D1. Low bitrate IMO is when you are closing in to 2000kps.

[jshumate]

I've been having a lot of issues with CCE hanging, both in native Windows XP on Boot Camp and in Parallels. I'm using QuickTime files as a source, so that could be the problem. Anyone have experience with this kind of thing?

Several disparate things can cause CCE to hang. On my old 1 GHz CPU, it would hang unless I changed the priority on it to RealTime. I don't recall it ever hanging after that. However, I have a 3200+ AMD CPU system and running CCE in RealTime mode makes no difference. It still hangs from time to time. I believe the problem is related to CPU temperature and if my PC gets too hot, CCE locks up. One final suggestion would be that you should NEVER use CCE to encode audio, so be sure to not do that as nothing good will come of that and you will only potentially cause problems for yourself.

[kumi]

I'm using QuickTime files as a source, so that could be the problem.

I seem to recall that SP2 includes a lot of Quicktime-specific improvements. Have you tried SP2?

[PhillipWyllie]

You should find that an average of 4600kbps is sufficient(as long as the whole movie isn't fast motion). Make sure when doing your tests to do a multipass encode as just one pass isn't a true reflection of how the output video will be. To squeeze more you could encode at 704*576( I find when viewing at this frame size there is little or no difference to 720*576). If you could provide a small example of your source I'd could help better.