combining Teegedeck's presets with Kopernikus's HVS ?

[ve1itas]

I am new to encoding and have been playing with constant quality (not quantizer) encoding. From what I understand, there seems to be two ways at the moment:
1. Teegedeck's presets: http://forum.doom9.org/showthread.php?t=119399
2. Eruendil's & Kopernikus's HVS: http://forum.doom9.org/showthread.php?t=114811

With Teegedeck's presets, it seems higher quantizer can be used for really bright or dark blocks -- compress better without losing perceived quality. Kopernikus's HVS limits modes for motion estimation (http://xvid.ist-dein-freund.de/short.html). Would it make sense to combine the strength of the two methods by doing a two pass encoding with Teegedeck's presets and activates Kopernikus's HVS in the second pass?

I've tried this with a small clip and the size goes down from 28.1 MB to 26.6 MB -- ~5% saving. Adapting from Eruendil's formula, I just add -hvs_aq "39 lc_lum - theta 49 lc_lum - theta + lc_lum 140 - theta + lc_lum 150 - theta + quant +" to pass2 command line.

Since I am a newbie, I figured I'd better ask the experts here. :)

added:
Also, would your opinions differ, if I can only use MPEG QM to be compatible with my standalone player.

[Sharktooth]

no since the Teegedeck presets use quantization matrices.

[henryho_hk]

Indeed, I am wondering how the VAQ patch and the HVS formulae will interfere with each other.

[Sharktooth]

im incline to remove the presets from megui since ppl is using VAQ and presets with CQMs (they modify the presets)... which is plain wrong.

[henryho_hk]

Teegedeck's presets are meant for their own specific CQMs. I don't think VAQ itself will have any (bad) interference with the presets (except that we can now remove the old AQ setting for >30% Preset). Instead, the real problem is that ppls often choose the presets subjectively, while they should do a compressibility test.

... "they modify the presets" ... are you referring to me? ~lol~

[Sharktooth]

it has, coz it basically does what CQMs do.
im referring to ppl in general.
btw, i've removed them.

[henryho_hk]

so.. do you mean that we should use H.263 or MPEG only for VAQ?

[elguaxo]

so.. do you mean that we should use H.263 or MPEG only for VAQ?

I think Dark Shikari mentioned that high bitrates CQMs (like 6of9) or above are not suited for VAQ.

http://forum.doom9.org/showthread.php?p=1121413#post1121413

I've tried VAQ only on low bitrate CQMs and it works great.

[Sharktooth]

try with h.263 it works evren better.
MPEG matrix is not an option.

[henryho_hk]

So, roughly speaking, we should use VAQ for low bitrate encodes (& "softening" CQMs), right?

[Sharktooth]

i didnt say that. i suggested to NOT use CQMs with VAQ. Read as VAQ should work better with standard h.263 quantization.

[ve1itas]

After reading more, I realized what I was trying to do in the first post is called Adaptive Quantization (AQ). :o As Sharktooth has suggested, using AQ with Teegedeck's presets is nonsense because both are doing the same thing.

However, Teegedeck's presets don't work with my standalone player, so I am using a modified preset which uses MPEG QM instead of the custom one. With this modified preset (from TGHQ-45), using the code in the first post seems to help reducing file size without losing perceived quality.

I've also tried using VAQ (http://forum.doom9.org/showthread.php?t=135093) with the modified preset with both MPEG and H263 QM. The quality loss seems to be more visible.

[Sharktooth]

However, i would try some experiments. Maybe you find some good combinations.

[henryho_hk]

I believe that the whole concept of Teegedeck's presets is to use the right settings for different materials. H.263+VAQ may replace the ">30%" or even becomes the new "<45%" Preset, who knows? My batch script is simply a platform to explore these opportunities (I hope it's not making things more complicated than it should be).

[Video Master]

AQ has worked gr8 for me with H.263 , Jawor 1CD , EQM v3 ULR rev 3 , EQM v3 LR for me [But with QPEL,Trellis disabled]
Havent tried VAQ though.

But for High Bitrate Matrices like 6of9 and EQM v3 HR the AQ hasnt worked properly for me.

[Everyone might have different views]

[elguaxo]

Trellis disabled

why ?

[Sharktooth]

trellis implementation in xvid is not as good as the x264 one...
the standard xvid AQ was made to rise the compression at expense of quality, so it is not a high bitrate option.
VAQ, on the other side, is always good but the mpeg4 ASP quant distribution doesnt permit to unleash the full power of VAQ, but it should still be much better than standard xvid AQ...

[henryho_hk]

Um.... Though not as good as x264, I do think it's better than none.