Usefullness of merange. 8 16 or 32 ?

[IgorC]

I did some short test with harry potter 2 trailer and another videos :

merange 8
merange 16
merange 32

And the results didn't different too much.

merange 8 comparing to me-e 16 only -0.01 -0.02 SSIM
merange 32 comparing to me-e 16 had - 0.01 of PSNR and SSIM.

merange 8 has 147% merange16's speed.

Maybe my tests were too short.
I'm doing right now more test. Can anybody test it too?

[Oline 61]

The vast majority of tests show that umh and 24 produce the best metrics (but soooo slow). 32 often reduces PSNR. ESA is too slow to be of any use at all. It's best to stick with 16 and hex or umh.

[IgorC]

Yes metrics are higher . No doubt. But is it really makin sence +0.02 ssim paying with low speed?

[Oline 61]

That 0.02 isn't worth the extra time it takes IMHO.

[IgorC]

i.e.
merange 8 - ssim 84.93
merange 16 - ssim 84.95

Something less than 0.15-0.2 ssim isn't noticeble.

[foxyshadis]

Are you testing me hex or umh? Assuming umh, are other options max quality, max speed, or a halfway tradeoff? That can make a difference.

Also, is this an average of multiple sources, or just a single random source? A higher me range would help more with faster-moving sources, or certain scenes in an otherwise sedate show, but how much I couldn't say.

[IgorC]

Yes, I'm testing with a heavy sets

x512.exe --pass 1 --bitrate 600 --no-dct-decimate --stats ".stats" --ref 16 --mixed-refs --no-fast-pskip --bframes 3 --b-rdo --bime --weightb --filter -1,-1 --subme 7 --trellis 2 --analyse all --8x8dct --qcomp 0.75 --me umh --merange 8 --progress --b-pyramid --direct auto --output 1x512_heavy_me8.mp4 harry.avs

Maybe subme7, trell 2 and a lot of RDO sets etc. are already doing a great part of job and for merange there is not much to do.

Video is Harry Potter Trailer - mixed high/low motions, darkness for weight test etc.

Going to test with subme 5/6 , hex etc.

[IgorC]

For folowing setings the difference between merange 8 and 16 is also only -0.01 opsnr.

This time sets aren't too heavy as before : now it is subme6 , ref 8, trell 1, hex.

x512.exe --pass 1 --bitrate 600 --no-dct-decimate --stats ".stats" --ref 8 --mixed-refs --no-fast-pskip --bframes 3 --b-rdo --bime --weightb --filter -1,-1 --subme 6 --trellis 1 --analyse all --8x8dct --qcomp 0.75 --me hex --merange 16 --progress --b-pyramid --direct auto --output 1x512_heavy_me16.mp4 harry.avs

[DryFire]

how much of an impact would you expect the video's resolution to have?

[SCIF]

I tested 8, 12, 16, 32 merange's at anime "Kenfu Denki Berserk" (source very cleanly 496x368 (for 640x480 - same result), 10001 frame).---qp 20 --ref 8 --mixed-refs --no-fast-pskip --bframes 3 --b-pyramid --b-rdo --bime --weightb --direct auto --subme 6 --trellis 1 --analyse all --8x8dct --me umh --progress --no-psnr --output The results was same as IgorC: high merange's it's not useful - gain in size +-15kb at 25.5Mb. Sorry for my english.

[shon3i]

You want to say that merange 8 have big gain in speed but noticable disturb quality

[SCIF]

You want to say that merange 8 have big gain in speed but noticable disturb quality
Do you can to do other conclusion from ower (not consistent) test??


I just finished my small encoding test. I coded 5:50 part of licensed dvd-9 "Chicago" - very clean source, 720x544. --pass 2 --bitrate 700 --stats ".stats" --ref 8 --mixed-refs --no-fast-pskip --bframes 3 --b-pyramid --b-rdo --bime --weightb --direct auto --filter -2,-2 --subme 7 --trellis 1 --analyse all --8x8dct --me umh --progress --no-psnr --output
merange16(30 750 689) - 77.16,
merange8(30 747 018 ) - 77.14.

[foxyshadis]

Testing how it fares on 8 fps low-bitrate video, to see if anything changes, and I see very marginal SSIM improvements at about half-D1.
8->16: .001 to .006 increases.
16->32: .000 to .003 increases.
Really, nothing to write home about, especially since each doubling of me range dropped by 1 fps (11->10->8.5). At full size, the differences were a little more pronounced, but rarely more than .01; the 16 did look a little better at that bitrate and size though. (Note, this is probably unscaled SSIM, so even multiplying by 100 it might not match yours. I'm unsure how MSU's works exactly.)

Command-line used:

"C:\Program Files\music-video\encoding\MeGUI\x264" --pass 2 --bitrate 75 --stats "Episode_01_final-chibi.stats" --keyint 400 --ref 6 --mixed-refs --no-fast-pskip --bframes 10 --b-pyramid --b-rdo --bime --weightb --direct auto --filter -1,-1 --subme 6 --trellis 2 --analyse all --8x8dct --me umh --merange 8 --threads 2 --thread-input --progress --no-dct-decimate --no-psnr --output "Episode_01_final me8 b75.mkv" "Episode_01_final.avs" --thread-input --aq-tcplx 8

(Yes, 75.)

[SCIF]

In this (http://forum.doom9.org/showthread.php?t=96636) thread Manao sad, that merange works only with --esa. With --dia, --umh it only slowdown without any effect. I offer rewrite contexthelp.xml in megui: <Recomended>8 to 16</Recomended> in section about merange.

[shon3i]

In this (http://forum.doom9.org/showthread.php?t=96636) thread Manao sad, that merange works only with --esa. With --dia, --umh it only slowdown without any effect. I offer rewrite contexthelp.xml in megui: <Recomended>8 to 16</Recomended> in section about merange.
what about --hex

[SCIF]

what about --hex
My tests show same about --hex.

[akupenguin]

merange applies to umh and esa, not dia or hex.
If you measure a slowdown from increasing merange on dia or hex, then it's time to rethink your benchmarking method, because you obviously can't distinguish between real and placebo effects.

[Prettz]

The vast majority of tests show that umh and 24 produce the best metrics (but soooo slow). 32 often reduces PSNR. ESA is too slow to be of any use at all. It's best to stick with 16 and hex or umh.
Measuring PSNR is a waste of time, it means nothing at all. Lowering the PSNR tells you nothing about how the quality of the picture has changed (within reasonable amounts, mind you).

Unless penguin says otherwise here, increasing the ME range should always increase quality slightly, if it has any effect at all for a specific encode. I would think there wouldn't be a single MB in your encode that looks worse using 32 than it did using 24 (unless the metric the codec uses to compare quality happens to fail for that instance).

[SCIF]

merange applies to umh and esa, not dia or hex.??? What is it (http://forum.doom9.org/showthread.php?p=825479#post825479)??
Unless penguin says otherwise here, increasing the ME range should always increase quality slightly, if it has any effect at all for a specific encode.
I think very very small part of people coding with -esa. I think most coding with -umh or -hex.
I offer rewrite contexthelp.xml in megui: <Recomended>8 to 16</Recomended> in section about merange.

[unmei]

??? What is it??
"It" is UMH..