x264 psy optimizations blind test

[MasterNobody]

Hi all. I am going to ask you to participate in this small visual testing. Your task is too choose the sample which you prefer the most from low bitrate samples and write that number. If you have problems to choose between two than write both but still try to rank them as first and second place.
Encoded samples (http://www.mediafire.com/?b11jym3obv4df36)
Original lossless video (http://www.mediafire.com/?99gsnf4vbqfypcl)

1. Don't look at any of the responses below this post until after posting your rating.

2. Feel free to post specific comments about specific files. While just knowing best is useful, it's most helpful if you do the full ranking.

3. You can analyze things in any way you want, but say how you did it. For example, if you just watched them all and then ranked them without freezeframing, say so. You could also do a rank of visual quality from watching, and then a second rank from freezeframing, if you wanted.

Note that the source is supposed to have grain here, so don't be silly and complain about the source material. Part of the test is grain retention. Of course, if you don't like the way in which the encoder retained the grain, that's a valid complaint.

[AlekseiV]

I'm lazy so I'm not testing these (yet, anyway) but I've mirrored the two files to avoid MediaFire.
Encoded samples (http://sovietpride.su/d9/samples.zip)
Original lossless (http://sovietpride.su/d9/BlackPearl.mkv)

[Sharktooth]

at a first sight there are 2 or 3 samples i like more than the other (3, 5 and 6) however, ill have a better look tonight and will post a full ranking.

[julius666]

i just watched them all and then ranked them without freezeframing :)

for me the top 3 is 5,1,4, in that order.

[Blue_MiSfit]

To me, 5 is subjectively the least annoying - at least in a quick roundup using fullscreen playback on a 17" 1920x1200 panel (no freeze frames).

[7ekno]

You mean despite the obvious Aspect Ratio error?!? Your DAR is 2.05:1, I suspect the correct one for this movie is 2.35:1 ...

I would order "quality" as:
5 - stable background with decent enough detail, but still a few halos
2 - stable background, a little detail washed out and still a few halos
1 - background becoming unstable, a little detail washed out with lottsa halos
7 - semi unstable background with alot of detail washed out with moderate movement artifact
4 - background unstable with flicker, more detail washed out and a lot of motion artifacts / halos
3 - background unstable with washed out detail, halos and motion artifact
6 - unstable background with alot of flicker with some washed out detail and lots of motion artifacts

If trying to lower bitrate, I would suggest running the clip thru a script like:


Source() # whatever is appropriate source
Crop() # For propper aspect ratio
MCTemporalDenoise(settings="medium", enhance=true, edgeclean=true, stabilize=true, screenW=704, screenH=352) #Change screenW and screenH for propper AR)


This will increase compression, stabalize the background, reduce halos and retain more detail ;)

7ek

[audyovydeo]

plain watching, no freeze framing.

1 and 5 were my perceived "best", I could distinguish a "worse" from the remaining lot.

cheers
a/v

[Sharc]

From best to worst:
1, 5, 2, 7, 6, 3, playback on PC.
Rating may be different when watched on LCD TV or with projector.

[doom_zy]

I think 2.mkv is the best,the frame look like more smooth.
for order
the better 2,5,7
and next 1,3,4,6

[AnonCrow]

Agree with most that samples 1 and 2 are near the top, but dislike sample 5.
Of course , even the source video has quite horrible artifacts (not talking about the grain), eg. around the hanging swords in the panning scene at around 0:03 to 0:07

[Sharktooth]

psy-optimizations are not about removing the grain... it's about KEEPING details (grain included since it's in the source clip)
if you want to remove the grain just use a smoother/denoiser.

[IgorC]

Realtime playback (no pauses)

Good results:
1 and 5 - both are good.
6 - hm pleaseant and something interesting. Might work. I find it the most detailed but a little bit annoying on the part of grain's artifacts.

Inferior results:
2 and 7 - Blur
3 - Ringing
4 - Artifacts

[Emulgator]

1 and 5.
As L3.1 and resolution seem to aim at smaller playback devices,
mainly watched in source resolution 720x352 on PC-TFT.
One third of the time in fullscreen on a 1920x1200 display.
Player: MPC, no adjustments.

[Sharktooth]

Im still undecided...

[doom_zy]

the final conclusion?which is the best?and the parameters?

[LaTo]

To me 2 and 7 are the best. (3 and 6 are the worst)

[Marius-the-Mad]

I'm still undecided. I'll have to analyse them closely this night...

I'll edit the post in twelve hours or so. [done]

****


Jesus. Grading this pack was hell. In terms of hard work, the previous test was peanuts compared to this one. I hope I have finally made my mind up about these (I'm still not sure whether I'm sure... :)).



My own, subjective results are: 5-1-6-4-7-2-3.



5: The best. Good backgrounds, not much of detail pumping, not much ringing, some grain retention. The best all arounder.

1: Good, but too unstable.

6: More grain, more ringing, some blocking or shimmering - but still quite good.

4: Backgrounds too unstable - sometimes good, sometimes bad.

7: Too much detail pumping due to the grain retention not being uniform enough (backgrounds).

2: Worse backgrounds, details washed-out here and there.

3: Too much blocking and ringing.



Viewing method: just playing the samples, one at a time, looped. No frame freezing, but a lot of timeline jumping.



Viewing environment:

- 24″ 1920*1200 H-IPS monitor

- night (too much light during the day for this kind of work)

- ffdshow outputting YUY2 -> DirectShowSource -> Lanczos4Resize(1920,768) -> Haali Renderer and its color space conversion (forced BT.601).


I hope it helps. Good luck with the tuning, guys, and :thanks: .

[jsquare]

No offense here, aside from the incorrect AR they'll look bad to me, and can't tell much of a difference. Maybe these clips were meant for an low-res/small screen and I'm testing them on a 37" 1080p LCD monitor where you can see every single defect.

[burfadel]

No offense here, aside from the incorrect AR they'll look bad to me, and can't tell much of a difference. Maybe these clips were meant for an low-res/small screen and I'm testing them on a 37" 1080p LCD monitor where you can see every single defect.

The purpose of the test isn't to compare visually transparent encodes, that would be a complete waste of time! By seeing every single defect you are in a prime position to choose the encodes that are visually best in your opinion, and then scale them to the worst :) having a result where the encodes are scaled from those that look bad to those that look just plain crap is still a result! :)

[infoeater]

Could you tell us what are differences between technology used in encoding samples?

[obieobieobie]

I prefer 1. No freeze frame.

[nm]

Could you tell us which settings have you used? Are settings in compressed samples manipulated, or the main difference is preprocessing and CRF?

I'm guessing the test was about experimental psychovisual tunings within x264, as the thread title and poster's identity imply. CRF was adjusted to match file sizes.