Go Back   nV News Forums > Linux Support Forums > NVIDIA Linux

Newegg Daily Deals

Reply
 
Thread Tools
Old 04-21-10, 06:32 PM   #1
Stu-e
Registered User
 
Join Date: Apr 2010
Posts: 9
Default VDPAU 2:2 pulldown detection failure

I have an H264 video clip containing progressive video (1080@50i) encoded as interlaced field pairs (2:2 pulldown).
When viewed with either XBMC or Mplayer with advanced deinterlace combined with inverse telecine, it is evident that deinterlacing is still taking place (line twittering on very small detail). In fact the picture is identical if viewed with inverse telecine disabled.

Am I right in believing inverse telecine should detect 2:2 pulldown and automatically disable deinterlacing?

I am using a GT220 so it should have enough power to do this.

Stu-e
Attached Files
File Type: gz 2_2ivtc_problem.tar.gz (52.5 KB, 99 views)
Stu-e is offline   Reply With Quote
Old 04-22-10, 01:56 PM   #2
Stephen Warren
Moderator
 
Stephen Warren's Avatar
 
Join Date: Aug 2005
Posts: 1,327
Default Re: VDPAU 2:2 pulldown detection failure

Stu-e, could you please upload a sample that demonstrates the problem to the NVIDIA file drop; see the sticky posts. Thanks.
Stephen Warren is offline   Reply With Quote
Old 04-24-10, 03:03 AM   #3
Stu-e
Registered User
 
Join Date: Apr 2010
Posts: 9
Default Re: VDPAU 2:2 pulldown detection failure

Hi Stephen
I have posted the clip at https://partners.ftp.nvidia.com/ as you requested.

The filename is
2_2_pulldown_Radnyclip2.ts

The video clip shows a close-up of a brass musical instrument. Look closely at the scratches on the surface of the instrument. With deinterlacing disabled the scratches and imperfections should appear stable and solid. With deinterlacing enabled they shimmer and twitter.

Stu-e
Stu-e is offline   Reply With Quote
Old 04-27-10, 04:55 AM   #4
Stu-e
Registered User
 
Join Date: Apr 2010
Posts: 9
Default Re: VDPAU 2:2 pulldown detection failure

Bumping
Stu-e is offline   Reply With Quote
Old 04-27-10, 11:30 AM   #5
Stephen Warren
Moderator
 
Stephen Warren's Avatar
 
Join Date: Aug 2005
Posts: 1,327
Default Re: VDPAU 2:2 pulldown detection failure

I've downloaded the file and see the issue, but haven't had a chance to investigate it yet. This is unlikely something that'll get fixed in just a couple of days.
Stephen Warren is offline   Reply With Quote
Old 04-27-10, 12:41 PM   #6
Stu-e
Registered User
 
Join Date: Apr 2010
Posts: 9
Default Re: VDPAU 2:2 pulldown detection failure

Just confirmation of the problem is enough for me.

Thanks for your response.

Stu-e
Stu-e is offline   Reply With Quote
Old 04-29-10, 10:38 AM   #7
Stephen Warren
Moderator
 
Stephen Warren's Avatar
 
Join Date: Aug 2005
Posts: 1,327
Default Re: VDPAU 2:2 pulldown detection failure

Stu-e,

Unfortunately, I don't believe we can make any improvements to this clip.

If the content you're playing truly is well-authored 2:2 progressive, with no interlaced material spliced in or field-level edits, then you may be better off just turning off the de-interlacer, since the material isn't really interlaced.

For explanation: There is little-to-no movement in the clip you provided. Hence, it is impossible to analyze differences between fields to determine where the frame boundaries are. Hence, it is impossible to determine the cadence (2:2, 3:2, ...) of the material. Hence, the inverse telecine algorithm doesn't activate, and instead we fall back to regular de-interlacing (i.e. mostly there is no difference between MPlayer command-line options "-vo vdpau:deint=4" and "-vo vdpau:deint=4:pullup" for this clip). For whatever reason, the de-interlacing algorithm doesn't recover the full resolution from the original image, and hence you see some slight twittering on the scratches. I briefly experimented with a few of ffmpeg's SW de-interlacing algorithms, and I believe I see the exact same kind of artifacts on this clip.

Last edited by Stephen Warren; 04-29-10 at 10:40 AM. Reason: Disabled smilies
Stephen Warren is offline   Reply With Quote
Old 04-29-10, 07:36 PM   #8
cehoyos
FFmpeg developer
 
Join Date: Jan 2009
Location: Vienna, Austria
Posts: 467
Default Re: VDPAU 2:2 pulldown detection failure

Quote:
Originally Posted by Stephen Warren View Post
I briefly experimented with a few of ffmpeg's SW de-interlacing algorithms, and I believe I see the exact same kind of artifacts on this clip.
You briefly experimented with a few of MPlayer's SW de-interlacing algorithms - someone is working atm at porting yadif to FFmpeg.

Do I understand you correctly that the auto-detection of progressive frames generally only works on 3:2 telecined input (with possible changes in the 3:2 pattern as produced by post-telecine edits) and not on constant 2:2 (progressive-looking) material?

Carl Eugen
cehoyos is offline   Reply With Quote

Old 04-30-10, 04:53 AM   #9
Stu-e
Registered User
 
Join Date: Apr 2010
Posts: 9
Default Re: VDPAU 2:2 pulldown detection failure

Ok
I can understand it is very difficult to detect interlacing where there is little to no movement. It is nonetheless a little disappointing. My interest stems from viewing live TV where material can change between interlaced and progressive from program to program, which makes it unrealistic to expect the viewer to turn deinterlacing on and off as required.

I suspect honoring any kind of stream flags that indicate progressive material would be more reliable.

I will try to find some more clips with more dynamic content to at least satisfy myself that 2:2 pulldown detection does operate.

edit:
If there is no movement surely it would be better to turn off deinterlacing?
Quote:
Hence, the inverse telecine algorithm doesn't activate, and instead we fall back to regular de-interlacing
Stu-e

Last edited by Stu-e; 04-30-10 at 05:00 AM. Reason: Afterthought
Stu-e is offline   Reply With Quote
Old 04-30-10, 12:07 PM   #10
Stephen Warren
Moderator
 
Stephen Warren's Avatar
 
Join Date: Aug 2005
Posts: 1,327
Default Re: VDPAU 2:2 pulldown detection failure

Quote:
Originally Posted by cehoyos View Post
Do I understand you correctly that the auto-detection of progressive frames generally only works on 3:2 telecined input (with possible changes in the 3:2 pattern as produced by post-telecine edits) and not on constant 2:2 (progressive-looking) material?
VDPAU's inverse telecine algorithm should work fine with 2:2 in general. The issue is that in this particular clip, there is little or no movement, and hence there's little to no difference between fields from different frames/pictures. Hence, it's impossible to measure/deduce the cadence. I imagine even if the clip were 3:2 pulldown etc., the same problem would occur.
Stephen Warren is offline   Reply With Quote
Old 05-01-10, 12:36 PM   #11
cehoyos
FFmpeg developer
 
Join Date: Jan 2009
Location: Vienna, Austria
Posts: 467
Default Re: VDPAU 2:2 pulldown detection failure

Quote:
Originally Posted by Stu-e View Post
I suspect honoring any kind of stream flags that indicate progressive material would be more reliable.
If you have any (real-world) sample where this would make sense, I will try to implement it.
(I suspect stations either send progressive flag for everything not truly encoded as fields - which never happens on MPEG-2 DVB afaik - or interlaced flag for both live and 2:2 content, but it would be nice if I'm wrong.)

Carl Eugen
cehoyos is offline   Reply With Quote
Old 05-04-10, 04:43 AM   #12
Stu-e
Registered User
 
Join Date: Apr 2010
Posts: 9
Default Re: VDPAU 2:2 pulldown detection failure

Cehoyos
Am I correct assuming progressive and interlace flagging reliability is generally quite poor?

Can you recommend any software tools freely available under linux for analysing dvb streams to check for the presence of such flags?

Are things more reliable for h264 streams? Reading around I get the impression there are no such flags. Instead field coding is implied from the h264 encoding method?

Stu-e
Stu-e is offline   Reply With Quote
Reply


Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


All times are GMT -5. The time now is 11:42 AM.


Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.