Some motion artifacts

[Caroliano]

I was tring some low bitrates encodes and notice some anoying motion artifacts in x264 build 284.

http://img110.imageshack.us/img110/6771/errorx2643dn.png

Is natural so much motion artifacts in this quality? I can make anything to avoid these artifacts or is a codec/bitrate issue?

And these type of small dots left behind don't apear in Xvid, even in his q31.

I have also a 102k example .avi video that show much better this anoyinance. Where I can upload it? Shoud I abuse a little from imageshack?

[Haze_NZ]

Firstly, post a bitmap of the framegrab rather than JPEG as you recompress the image using JPEG, meaning it is not an accurate format for still image video comparisons. If you are willing to upload the video, a good place to go to is www.rapidshare.de . It is the perfect site to upload to for your purpose....

[CiNcH]

@ Caroliano,

please have a look at my sample clip (~4.5 MByte) and tell me if your artefacts look the same.

http://members.aon.at/cinch/artefacts.mp4

They appear roughly at second 18 (ignore blocking in the blue sky) to the right shoulder of the Anubis warrior.

[Caroliano]

Video uploaded: http://rapidshare.de/files/4016650/Error_x264.avi.html

@Haze NZ :thanks for the tip.

I also re-uploaded the image to an .png but I think it don't helps so much since we have now the video that is far better to see these artifacts.

@CiNcH : Since I instaled the new 2005/08/03 version of FFDShow I can't see some x264 clips and your is one of them... I seemed it by VLC and my first clip artefacts are realy like the ones coming from the big torch.

A time ago I already noticed these dots left behind in a medium bitrate encode too. I thought it was because a bad misture of Xvid's ringing and x264 in-loop filter. Here a 1.43 MB sample with audio: http://rapidshare.de/files/4018050/error2.avi.html

Look near of her hair and her blue dress. There is "alot" of small black points after she moves... I circulated some small points in this picture (http://img110.imageshack.us/img110/3752/error26mx.png). Also you can see bigger points in her mother skirt and when she goes out for the first time.

[CiNcH]

So the same artefacts...

These artefacts are ringing artefacts. As you probably know ringing appears at sharp edges. When encoding with x264 at too low bitrates ringing is taken to the next frames via motion estimation. When those sharp edges move this effect becomes apparent.

Here's the explanation by akupenguin:

If you perform motion compensation and there are artifacts in the source frame, then those artifacts will be propogated. They only disappear when either (1) new data appears and covers them, (2) enough time passes that they get blurred away by multiple iterations of motion interpolation, or (3) a future frame gets low enough quantizer to fix the artifacts. Note that a sufficiently smart encoder could decide when to invoke (3), and we have discussed possible algorithms, but don't expect it in x264 soon.

Still I have never seen such behaviour with any other codec.

[Caroliano]

These artefacts are ringing artefacts. As you probably know ringing appears at sharp edges. When encoding with x264 at too low bitrates ringing is taken to the next frames via motion estimation.
The strange is that they are black... The ringing I know is not black... And a smart encoder shoud not use the ringing in the motion estimation... I think that the problem could be there.
enough time passes that they get blurred away by multiple iterations of motion interpolation
I confirm. This ocurrs alot in my 2° example that is wich a higher quantizer (24~26 for p-frames). In the first they only disapear when crash in something...
(..) a future frame gets low enough quantizer to fix the artifacts
I don't think that a strategic low quantizer frame would help because the black points keep apearing all the time... They should not be created by the way.

[DeathTheSheep]

Well, yeah, I suppose that's kinda ugly, seeing artifacts at this bitrate. But overall, the quality beats anything else you could get at this rate (except, of course, the new VP7).

But when I re-encoded your little segment in Q=37, I didn't notice all too many bold artifacts....especially for re-encoding with my new, super-low bitrate. The only wierd things were the blocks flashing in the background where motion had occured just prior.

Notice how the pantry/cabinet behind the Mom's head starts to shimmer for no reason at all? Now THAT is because of B-frames+motion artifacts+re-encoding, if I may so predict (a 10-second prediction).

Here's the direct link. (DONT EVER USE RAPIDSHARE!! Use www.megaupload.com or www.nerdswap.com, www.savefile.com etc. :p)
http://filepost.us/userfiles/xsquaredx/AIR-ROAR.avi

[Caroliano]

@Cinth : Why you re-encoded? And these shimering for no reason for me is because a too high in-loop filter. Re-encoding my source, that don't have b-frames, with deblock +6,+6 I get these shimerings in other places.
And a smart encoder shoud not use the ringing in the motion estimation...
Now thinking about it again the motion estimation shoud use the ringing in the motion estimation carring thogether with the sharp edge that made they apear.

I think that it shoud have an high priority because, to fix this motion error, the codec certanly waste bitrate, so the encoding efficiency is afeted. And, of course, because they are anoying in lower bitrates...