b-frame ghosting (d)effect

[Omni]

i just made some encoding tests when i noticed that i get some light ghosting effects . if first thought it's because of some filters i used. but encoding using p-frames only removes the ghosting (i tried the last 3 xvid releases with the same result). normally it's not obviously visible and only occures on certain scenes. i can upload some screenshots if you don't already know this problem and have a solution for me. (well i used the search button^^ but didn't find anything that matches my problem)

[Koepi]

This sounds very strange - so yes please, upload not only a screenshot, but a small snippet (avi) from the source. I wonder why noone else experienced that yet...

Regards
Koepi

[Omni]

well, i just noticed that i'm unable to access my t-online webspace right now. i'll upload it tonight on my dyndns webspace or if an admin grants me to post attachments i'd post them right in here.
thanks for your fast response and concern in this matter :)

[Omni]

i made some screenshots (i can give further examples if you want but i hope this is enough for the moment)

1. scene P-frames only:
http://www.sheridan.privat.t-online.de/01_p.jpg

1. scene 4 consec. B-frames @+20sens (@0 it's the same)
http://www.sheridan.privat.t-online.de/01_b.jpg

2. scene P-frames only:
http://www.sheridan.privat.t-online.de/02_p.jpg

2. scene 4 consec. B-frames @+20sens
http://www.sheridan.privat.t-online.de/02_b.jpg


example movies:
Movie without B-frames (http://www.sheridan.privat.t-online.de/p_only.avi)
Movie with B-frames (http://www.sheridan.privat.t-online.de/4b_20bsens.avi)


those effects are visible in various scenes, i just noticed them while tuning my hq-custom matrix.
ghosting only appears with b-frames and is independent from the used matrix.
it's more visible if the picture is denoised.
to make sure that it's not a filter or virtualdubmod related problem i saved the clip as rgb uncompressed.


edit, i forgot to give my complete settings:
source: DVD
dvd2aviidg
avisynth 2.53
############################################
mpeg2source("temp.d2v")
Telecide(guide=1,blend=false,reverse=false)
Decimate(cycle=5)
LanczosResize(640,480,8,0,704,480)
############################################
Virtualdubmod with filters: tempsmooth+sshiq
XVID Settings:
Singlepass (quant=2)

Profile: urestricted
Quantisation type: MPEG
Enabled:
QPel
B-VOPs: 4 / 1.50 / 1 PBS+Closed GOV

Advanced Settings
Motion search precision 6
VHQ 4
Use Chroma (chrome enhances the effect in certain cases)
Trellis quant enabled

Zoneoptions->B-frame sensitivity set to 20

[MfA]

Frame 34&36 are better examples IMO.

[Omni]

well frame 36 shows nice compression artefakts. i think this corruption ist 50% because p-quants are at 2 and b-quants at 4 and the other 50% may be from ghosting problems. but normally a b-frame quant shouldn't be that obious or am i wrong?

[ObiKenobi]

I don't see any ghosting, as in a remnants of a past frame in the frame preceding, but I do notice alot of noise around black lines. Is this what you were referring to? Because that is nothing more than an mpeg artifact and would be caused by too little bitrate for those frames. So, if you want to try to fix this, try decreasing the quantizer ratio for b-frames to see if this provides better results. B-frames will always be more compressed then p-frames and can result in the kind of noise you are seeing.

[Caspar]

Same here... unless there's something wrong with my eyes, I don't see any ghosting from the screen shots. I even copy and pasted the screen shots one on top of the other in paint, can't see ghosting.

[ObiKenobi]

Well it's good to know it's not just me. :)

[Omni]

take a look on fileset 2 (better look at the avi using zoomplayer or vdub watch it frame by frame) the lineartifakts on the middle finger are exakt where the nextframe has the fingeroutline so there's data in it which should appear a frame later.
my settings were p=2 and b=4 which should be okay. if i only use p=4 without b-frames the artifacts are gone. if i reduce the b-frame quant to 3 than the artifacts are a bit less visible but still there. so it seems that b-frame can cause ghosts of the next frame. a high p-frame quant can cause artifacts of a previous frame.
b-frames really help to save some space but isn't there a way to prevent such ghosts/artifacts? maybe some tuningoptions for b-frames (despite their quant).

[RadicalEd]

Are you sure that's not just the result of cartoon mode? Looks more like that than any b-frame problem I could imagine.

[ObiKenobi]

Well looking at it again using vdbu this time I see some very very faint ghosting on both versions, but only when single framing at 300% is it really visible. Playing at normal speeds it's unnoticeable. Seems to be more an artifact from the IVTC rather than something relating to the use of b-frames. But considering its only really visible when single framing I wouldn't consider this any big deal.

[Omni]

Originally posted by RadicalEd
Are you sure that's not just the result of cartoon mode? Looks more like that than any b-frame problem I could imagine.

i didn't use cartoon mode. i made about 30 samples now and there are two things in common:
these errors are only on b-frame encodes visible and their strength is related to chroma motion detection.



Well looking at it again using vdbu this time I see some very very faint ghosting on both versions, but only when single framing at 300% is it really visible. Playing at normal speeds it's unnoticeable. Seems to be more an artifact from the IVTC rather than something relating to the use of b-frames. But considering its only really visible when single framing I wouldn't consider this any big deal.


well the first 3 frames in the right corner it's visible and there are more scenes where it is visible. most errors occure in dark areas.
it's definitly not a IVTC problem since (read my old posts) i tested it with an rgb avi (uncompressed) output as reference so the error definitly lies in xvid and definitly is related to b-frames :D
i'm using xvid for over a year now and well, i use it for highquality encodes and i never really noticed this bug since i normally don't encode anime dvds. the only other problem i had and have is that using more b-frames (more than 1 consequ) makes the playback a bit jerky on my AXP@2,3ghz (well hard to notice for some people but some noticed it when i gave them a sample) it sometimes looks like a frame is dropped which isn't the case...
okay back to the topic.
up till now i said everything i know concerning my little problem. i'll try to reproduce it with other sources but in the end it's koepi's decision wether to investigate this problem further, or not.
maybe i'm to strict concerning quality. but for me xvid is a highquality solution. i like to use it with my own hq matrix and b-frames but more i don't like messy frames :-S
hopefully there is a way to fix this without leaving out b-frames or setting a zone for every problematic scene^_^

[ObiKenobi]

Well considering I could see ghosting in both versions, both with and without b-frames I'm still more inclined to think its more related to your IVTC and not the b-frames. But hopefully someone like Koepi and Nic can help you if indeed it is b-frames causing this. Here's hoping you find the answer to your problem. :)

[MfA]

In the scene scene where the kid is leaning back (with frame 34/36) there is an extra moving edge to the right of his hair (it does look like ghosting). Dunno if that is a ME problem, or if that is just the best way to encode it in a RD sense ... it doesnt look pretty, although if it hadnt been pointed out I wouldnt have noticed. Im not a very critical person as far as image quality is concerned though.

Look at the AVIs not the images, those were poorly picked.