low bitrate settings for xvid?

[HalfHuman]

hello again!
i have a cheapskate question
say i want to fit a 2.5 or even 3 hours of movie on a 700mb cd. what settings should i use with xvid and what should i never use?:confused:
i read lots of stuff on ur forum and others. i have a general ideea but i want to hear it from u again. i must have missed the subject or u haven't talked about it. (maybe because of the unusual request)
so i would like to know whether i shoud use qpel, chroma optimizer, adaptive quant, how much consecutive b-frames, what matrix, what kind of sound, what kind of movie container, what avisynth filters etc. the goal is to achieve the maximum quality within the given filesize. it is important only to fit the requested (small) filesize and QUALITY (in the given order).
thanks in advance!
:)

[Ark]

What source is this movie? And what aspect ratio it has?
And what's the target (PC or TV)?

700mb for 2.5/3 hours is very low bitrate, so low resolution too.

for Avisynth
I suggest Bicubic(or even bilinear) resize, undot(), and a good denoiser like Convolution3d(), RemoveDirt(), MipSmooth(), Temporalcleaner(), Vaguedenoiser().

for XviD
I suggest to use AQ, GMC, NO qpel, a matrix like H.263 or HVS-good, b-frames at default values, VHQ4 (for b-frames too), Trellis.

For resolution, if the source is well compressible (low motion, small noise, high contrast content) you can use 640xXXX, if results are bad, lower gradually resolution at steps of 16 (for horizontal resolution, adjust vertical accordingly but always mod-16 to avoid decoding problems).

The container depends on target, if it's PC, I suggest Matroska+Ogg vorbis Audio (or AAC). If it's TV (so standalone DVD-Player), for max compatibility use AVI+MP3 CBR.

[Teegedeck]

It depends on the resolution. If you lower it to a sane level (say, 512x) you could use SixOfNine-HVS matrix, I'd say, and keep QPel activated.

If you don't resize, you could try Javor's 1-CD matrix and should deactivate QPel. VHQ=4 really is a must here. Also adaptive quantization, chroma ME and trellis, leave the b-frames at their default settings. Chroma optimizer is nice, GMC would be good.

Try some mpeg2dec3-postprocessing if your source is MPEG-2. Start out with

cpu2="xxxxox"

Edit: For Avisynth filters, I'd only suggest unfilter() or removegrain/removedirt. If your source isn't really noisy.

For sound I'd use HE AAC which leaves the choice between Matroska or MP4 as containers.

Also try video-codecs like Real or VP and see whether their inbuilt deblocking serves your purpose.

Edit: Ark beat me to it. Very good suggestions, too.

[Sharktooth]

Originally posted by Teegedeck
For sound I'd use HE AAC which leaves the choice between Matroska or MP4 as containers.

... you forgot AVI :)

Also try video-codecs like Real or VP and see whether their inbuilt deblocking serves your purpose.

For such a low bitrate RV10 or VP62 is the way to go but HalfHuman just asked about Xvid, so there is almost nothing to add to the above suggestions by Ark and Teegedeck.

[pelle412]

If you have the time and patience use PixieDust (avisynth filter). It's ridiculously slow but the noise removal / quality and the compressibility gains are superb.

[Peter Cheat]

You didn't say whether you intend to encode anime or natural film. The best advice I can give is to find a typical small sample of about 3mins and try encoding with different filters and settings. I use PixieDust for low bitrate encodes too, that should improve compressibility. The H.263 is good for low bitrate encodes, it makes blocking less visible. Good luck.

[HalfHuman]

tnx for the quick answers!
to make things much clear:
- natural film
- destination PC
- resizing is not a problem (in this case is a must ):D
- as for codec i prefer xvid like most of ya
feel free to complete the sugestions and for the FAQ and documentation writers: u should include recomended settings for high/low bitrate, noisy/clean source, high/low res, anime/natural, target pc/tv etc. wow the plot thickens! there are already 5!=120 possibilities. :eek:
it should be short like the answers in this forum in a new section.
just a sugestion.
be well!

[Chainmax]

How about putting 300 minutes of low detail cartoons on a CD? I'd be using HE-AAC @ 48kbps (which should be resampled, right?), so the average video bitrate would be ~270kbps. Would H.263 still be a good choice or should another matrix be used? Resolution would be 320x240.

[Sharktooth]

Try my new EQM V3LR or ULR matrices.
I just need some feedback :)

[forodwaith]

Hi,
wanted to ask what the biggest disadvantages of Filters like Undot or Convolution3d are. Are they smoothing/ bluring the image very much?
I just ran a compressibility check on movie and got almost 30% more in the comp check just with the use of both filters.
Also gained another 20% with bicubic resizing.
Shouldn't i then not always use these filters with a higher res than i would without these filters?

Thanks in advance

[Teegedeck]

Encode the movie with and without filtering. Compare the results. Go for the version which looks better. And I bet it'll be the one without filters, except if a) the source actually was low quality or b) you were trying something quite extreme.

High resolution with noise-filtering regularly looks worse than slightly lower resolution without filtering. I would decide on a resolution first and on filtering-or-not-filtering next.

Undot is pretty harmless and I would use it for 1-CD encoding in most cases.
Personally, I don't resize at all, so no advice for me on resizers.

[Sharktooth]

The softer is the resize the higher will be the gain in compressibility but at the cost of sharpness.
However like Teegedeck, i dont like to resize too, but i prefer some light filtering (undot, vaguedenoiser, convolution3d ...) to gain some compressibility when needed.
Resize is a lossy process and filtering too. But some filters are also made to improve image quality (removing artifacts, noise, imperfections etc) and with so many filters to choice from you can manage to obtain much better results than with resizing.

[HalfHuman]

in my grand experience ;) i can tell u that filtering is always a lossy process whether is resizing, cleaning etc.
it's only logical that when u resize the image (shrink it) u end up with less information to compress (almost always) and that's why u get more compressibility. i rather like resizing bilin or bicubic rather then using a filter like unfilter(), convolution3d, peach. i prefer only using undot() and then resizing the picture (bilin for softer, bicub not so soft). denoisers make the movie more compressible but they usually take to much detail out for clean movies. some prefer the image a little blocky instead of a vectorial like image where u can see no wrinkles, moles and such things that make the movie look more real.
convolution3d never used it. no oppinion
hope i helped u

[Clock]

Could someone please explain, why it is recommended to use default b-frame-settings for very low bitrates?

I'm no pro, but from my understanding, b-frames "save space". Without trying to go into detail, it comes down to that: b-frame good for lowering bitrate, probably bad for quality.

But then there's this "save bits here, use them there" scheme. Which usually should improve overall quality.

So why not use more bframes for low quality? Whoever tries such low bitrates (as I sometimes do) knows, he won't get superior quality. So I often use a max of 5 bframes and also increase sensitivity. Where's the error in my thinking?

Clock

[Teegedeck]

Make a test. Encode your movie (or a clip from it) at the same quantizer with different b-frame settings. More b-frames will NOT in all cases result in a lower filesize. Also, gains from more b-frames get more and more miniscule the more b-frames you use; like 2 b-frames 30% savings, 3 b-frames 33% savings, 4 b-frames 33.5% savings... At the same time you increase the probability of quality glitches, and that may not be worth it. A similar thing goes for b-frame sensitivity. You are likely to loose something if you fiddle with the b-frame settings, so the general advice is not to do it.

[EthanoliX]

My recommendation:

max. 3 b-frames @ sens. 5 and q-offset 5 or even 6
anamorphic resolution: 400*288 instead of 512*288 for 16/9 movies
q-pel
GMC + VHQ4
low bitrate matrix
90 min. CDs, mode2 => 895 MB
ogg vorbis or better HE-AAC