--me tesa

[Asmodian]

I believe the information is this thread is still current for x264. Nothing about how motion vectors are handled in x264 has changed fundamentally.

Without testing your sources with different merange values I suggest not changing what the presets use either for 720 or 1080p. Remember the search range is not the maximum length of the vector. Preset placebo is tuned for maximum quality or minimum size with no thought for speed so if it is not used in placebo it will not, generally, help quality/size.

edit:
merage of 64 or higher is unlikely to find better mv candidates and can actually result in worse quality. 32 is the sweet spot

Shouldn't the preset plecebo be changed to use a merange of 32?

[microchip8]

I believe the information is this thread is still current for x264. Nothing about how motion vectors are handled in x264 has changed fundamentally.

Without testing your sources with different merange values I suggest not changing what the presets use either for 720 or 1080p. Remember the search range is not the maximum length of the vector. Preset placebo is tuned for maximum quality or minimum size with no thought for speed so if it is not used in placebo it will not, generally, help quality/size.

edit:


Shouldn't the preset plecebo be changed to use a merange of 32?

it won't be called placebo otherwise :P

What I said above comes from the mouth of Dark Shikari. I have no idea why placebo uses 64. Maybe just to max out everything, regardless whether it'll make a diff or not?

[sneaker_ger]

Placebo "only" uses 24.

[detmek]

it won't be called placebo otherwise :P

What I said above comes from the mouth of Dark Shikari. I have no idea why placebo uses 64. Maybe just to max out everything, regardless whether it'll make a diff or not?


Preset placebo uses 24, not 64. With that in mind one should probably not go higher then 24.

Edit: Sneak_ger was faster.;-) Also, I had to check if something might change in presets as I never used any preset slower then Slow.

[microchip8]

Placebo "only" uses 24.

hmm, indeed. I should look before speaking :)

[infoeater]

AFAIK 32 usually gives better results then 24, but 24 is sometimes better. As I guess 24 has been chosen for placebo to reduce occurrence of samples witch compress better on very slow than on placebo.

[kotuwa]

Well...
From this post bt , what I get is 64 can be a good value...
http://forum.doom9.org/showthread.php?p=1544428#post1544428
Extremely high merange (e.g. >64) is unlikely to find any new motion vectors that are useful, so it may very slightly decrease compression in some cases by picking motion vector deltas so large that they worsen prediction of future motion vectors in the rare cases they're locally useful, making them worse than useless.
The effect is so small as to be near-negligible, though, and you shouldn't be using such insane settings.

What it says is if it is higher than 64, then it might be slight disadvantage sometimes... right???


Preset placebo uses 24, not 64. With that in mind one should probably not go higher then 24.

Preset placebo is tuned for maximum quality or minimum size with no thought for speed so if it is not used in placebo it will not, generally, help quality/size.

As most would say, the sample encodes I done has not given any visible advantage by using placebo (or just subme=11) compared to Veryslow (or just subme=10)... (sometimes, placebo/subme 11 gives slightly bigger files, probably with a high quality, but can not be seen by our eyes)...

But merange=64 has showed a size advantage...
The file size under same CRF is little bit smaller than merange=24... In those samples.. It may differ in other samples!...
In that case... VerySlow with merange 64 gives smaller files and Placebo (with 24) gives bigger files under same CRF.... Quality would be high in placebo if measured, but not detected by the naked eye...

And I have seen that Full HD footage would be better with higher merange as it is a pixel number and it should be scaled with the resolution...
!?!

[infoeater]

@kotuwa
Short answer: For statistically best quality use --merange 64.

Following is summary from previous pages of this topic (nothing changed):

On this chart
http://akuvian.org/images/tesa2.png
you can see average results with --merange 4 8 12 16 24 32 48 64.
As you can see 48-64 give best results (difference between 48 and 64 is lower then inaccuracy of the test caused by selecting limited number of samples)

but for some samples 24 may be better:
http://akuvian.org/images/tesa-grimmc.png

which is overcompensated by better result with 64 in some other samples:
http://akuvian.org/images/tesa-hsing.png

Anyway difference is small.

[infoeater]

Maybe there is correlation between best --merange and video_resolution/fps, but I believe it is not linear or scalable in an easy way.

[detmek]

Same CRF value with different settings gives different quality and different file size. So, it can not be used to compare effects of various x264 settings. Only way to do that is to use 2-pass encoding mode at same bitrate/filesize.

If you want to compare --me tesa --merange 24 vs --me tesa --merange 64 use 2-pass encoding and set same bitrate for both encodings. Then, you can do a visual comparisson by directly comparing both encodes or you can and --tune ssim --ssim or --tune psnr --psnr to quantify that difference.

[Blue_MiSfit]

Guys, please don't bring old threads back to life!

[pcordes]

Since this thread is already reanimated, is it ok to post more?

My guess, based on no hard evidence, is that high me-range is most useful with low-fps or otherwise jerky video where objects move a lot of pixels between frames. Since motion search starts with the prediction vector from last frame, the jerkyness matters, not just speed.

You might find this in anime or handheld non-stabilized content.

You'd tell it was working by comparing SSIM at constant bitrate, and/or seeing a higher % of skip blocks in the stats output (https://en.wikibooks.org/wiki/MeGUI/x264_Settings/x264_Stats_Output).