Couple of newbie questions for Sharktooth

[Roscoe62]

I've been using Sharktooth's EQM V3LR matrix and have been very happy with the results I'm getting.

However, a couple of the files I've been converting with XviD did not fare very well when checking compressibility in Gordian Knot and in order to maintain a relatively small filesize I thought I'd try out the EQM V3ULR matrix.

I was shocked to see that while checking compressibility in GK using the ULR matrix, I'd end up with percentages well over 100%. In order to get down to the 65-70% ratio I normally use, I would need to significantly reduce the filesize - not that I'm unhappy to do that, but all the guides I've read recommend sticking to bits/pixel values of around .17. When I reduce the filesize to get the ratio down to a reasonable value I'm seeing the bits/pixel values get down to around .11 which has me a little nervous.

I'm only encoding individual TV episodes originally sourced from TV series DVD boxsets. Each episode is 4:3 aspect ratio and mainly PAL - (some NTSC) all around the 22 min mark and, being TV sourced, the PQ isn't that great to begin with. Some are cartoon based (Simpsons/Futurama/South Park) and some are not. I always use a 2 pass encode and add the PixieDust filter as I like the results that gives.

So far I've been aiming for filesizes (including a single VBR MP3 soundtrack) of around the 200Mb mark.....smaller is better - but only if PQ doesn't drop too much.

My questions are:

1) When using the EQM V3ULR matrix, is it NORMAL/ACCEPTABLE to see the bits/pixel value so low? Are there any guidelines here?

2) Is there a VAST difference in picture quality between the LR & ULR matrix encodes? (I know this is a subjective question, but a subjective answer from an expert is good enough for me!)

If the difference in PQ isn't a lot (given my source material) and it means a smaller filesize then I'd be happy to switch to ULR as I've got a TON of episodes to encode, and the HD on my server (where all these episodes are going) is starting to fill up!

Can anyone answer my questions?

[Sharktooth]

Ok, in general, low bitrate matrices are made expecially for those cases when compressibility is too low and you cant exceed a certain size.
Now, i dont exactly know how the GK compressibility test works (i love the old notepad -> avs -> virtualdub method) but, i think 100% refers to quant 2 (am i right?).
The problem is with ULR you get well over 100% that means you will waste bits for nothing and there is a risk ratecontrol will go nutz.
On the other hand the LR matrix does not compress very well...
The possibile solutions are 2:
Use the LR matrix and lower the resolution/apply heavier filters
or
Use the ULR matrix with some "workarounds" for compressibility.
Ok, the first solution is easy.
The second is more tricky. I'll explain...
The problem is it compresses too much, so disabling things that helps compressibility may help obtaining some more quality and the desired size. For example try disabling GMC, B-Frames, Adaptive Quant and stuff like that. Try using light filters etc... Repeat the compressibility test and look at the result.
If it's still over 100% then you can safely do a 1 pass encode at fixed quant 2 and obtain the best from that matrix. Size will be lower than what you intended though.
Now, let's answare your questions.
1) Low bitrate matrices will always have lower bits/pixel values due to their nature. The result is usually acceptable... they were made for that purpouse... Remember the bits/pixel is just an "indicative" number for an average compressibility. If it compresses more then the bits/pixel could be low.
2) It depends on the source and the bitrate choosen. If you do a Q2 encode with both matrices LR will look better but the final file size will be much bigger. On the other hand keeping the same filesize for both matrices ULR will use lower quantizers than LR and if it does not get near saturation it will look better.
Last thing, dont forget the basic rule: "more size, better quality".
But sometimes the differences are not so big...

[Roscoe62]

Sharktooth,

Thank you for such a comprehensive answer.

When I started converting the TV episodes to XviD's the average filesize was around 200-250Mb per episode. However in some series some of the episodes got up to around 300Mb which was a little too large for my taste. I've learnt I can control the filesize a bit better now, but the 200Mb average size was just a default based on my early experiments rather than having a specific filesize in mind. The bottom line for me is that if the episodes had the same PQ (or at least close) but with a smaller filesize then that would suit me perfectly.

I guess I'll just have to run some experiments to see how much of a difference it makes to PQ - one encoded with EQM V3LR & one with EQM V3ULR - do compressibility tests on both before beginning and aim for the same percentage rate, not filesize differences.

Once completed I'll watch both of them to decide whether I can live with the PQ difference. Remember, a lot of the TV stuff I do isn't great quality to begin with. A lot of source material-inherent defects can't be blamed on the encode or the matrix!

BTW....thank you for the effort you put into these matrices. I know I sure appreciate the results! :)

[Sharktooth]

Thanks.