Jittery motion

[luv2hike]

This one is hard to explain and my wife doesn't see it. But when I transcode an HD recording using an RF of 22, the picture quality looks perfect, however when watching the video, something just seems "off". It's almost like frames are being dropped? The motion of people's hands and heads just doesn't seem as smooth as the original. It has a robotic, jittery feel to me. Does that make sense? I'm very new at all this. Any ideas what parameters control this behavior? Here is my command line used and the run info:

Code: Select all

HandBrakeCLI -i "$1" -o "$2" -f mkv -e x264 -q 22 -E faac -B 192 -R 48 --strict-anamorphic -d slow -5 -8 medium

Code: Select all

HandBrake 0.9.5 (2011010300) - Linux i686 - http://handbrake.fr
2 CPUs detected
Opening /data/TVarchives/5081_20110110185900.mpg...
[15:25:49] hb_scan: path=/data/TVarchives/5081_20110110185900.mpg, title_index=1
libbluray/bdnav/index_parse.c:157: indx_parse(): error opening /data/TVarchives/5081_20110110185900.mpg/BDMV/index.bdmv
libbluray/bluray.c:960: nav_get_title_list(/data/TVarchives/5081_20110110185900.mpg) failed (0xb1bb9c0)
[15:25:49] bd: not a bd - trying as a stream/file instead
libdvdnav: Using dvdnav version 4.1.3
libdvdread: Using libdvdcss version 1.2.10 for DVD access
libdvdnav:DVDOpenFileUDF:UDFFindFile /VIDEO_TS/VIDEO_TS.IFO failed
libdvdnav:DVDOpenFileUDF:UDFFindFile /VIDEO_TS/VIDEO_TS.BUP failed
libdvdread: Can't open file VIDEO_TS.IFO.
libdvdnav: vm: failed to read VIDEO_TS.IFO
[15:25:50] dvd: not a dvd - trying as a stream/file instead
[15:25:50] file is MPEG DVD Program Stream
[15:25:51] add_audio_to_title: added AC3 audio stream 0x80bd
[15:25:51] scan: decoding previews for title 1
[15:25:51] scan: audio 0x80bd: AC-3, rate=48000Hz, bitrate=448000 Unknown (AC3) (5.1 ch)
Scanning title 1...
Scanning title 1...
Scanning title 1...
[15:25:52] scan: 10 previews, 1920x1088, 29.970 fps, autocrop = 0/8/0/0, aspect 1.76:1, PAR 1:1
[15:25:52] scan: title (0) job->width:1920, job->height:1072
[15:25:52] libhb: scan thread found 1 valid title(s)
+ title 1:
  + stream: /data/TVarchives/5081_20110110185900.mpg
  + duration: 01:07:15
  + size: 1920x1088, pixel aspect: 1/1, display aspect: 1.76, 29.970 fps
  + autocrop: 0/8/0/0
  + chapters:
    + 1: cells 0->0, 0 blocks, duration 01:07:15
  + audio tracks:
    + 1, Unknown (AC3) (5.1 ch) (iso639-2: und), 48000Hz, 448000bps
  + subtitle tracks:
    + 1, Closed Captions (iso639-2: und) (Text)(CC)
[15:25:52] 1 job(s) to process
[15:25:52] starting job
[15:25:52] sync: expecting 120980 video frames
[15:25:52] job configuration:
[15:25:52]  * source
[15:25:52]    + /data/TVarchives/5081_20110110185900.mpg
^MEncoding: task 1 of 1, 0.00 %[15:25:52]    + title 1, chapter(s) 1 to 1
[15:25:52]  * destination
[15:25:52]    + /data/TVarchives/5081_20110110185900.mkv
[15:25:53]    + container: Matroska (.mkv)
[15:25:53]  * video track
[15:25:53]    + decoder: mpeg2
[15:25:53]      + bitrate 20000 kbps
^MEncoding: task 1 of 1, 0.00 %[15:25:53]    + frame rate: same as source (around 29.970 fps)
[15:25:53]    + strict anamorphic
[15:25:53]      + storage dimensions: 1920 * 1088 -> 1920 * 1080, crop 0/8/0/0, mod 0
[15:25:53]      + pixel aspect ratio: 1 / 1
[15:25:53]      + display dimensions: 1920 * 1080
[15:25:53]    + filters
[15:25:53]      + Decomb (default settings)
^MEncoding: task 1 of 1, 0.00 %[15:25:53]      + Deinterlace (ffmpeg or yadif/mcdeint) (default settings)
[15:25:53]      + Denoise (hqdn3d) (default settings)
[15:25:53]    + encoder: x264
[15:25:53]      + quality: 22.00 (RF)
[15:25:53]  * audio track 0
[15:25:53]    + decoder: Unknown (AC3) (5.1 ch) (track 1, id 80bd)[15:25:53]      + bitrate: 448 kbps, samplerate: 48000 Hz
Encoding: task 1 of 1, 0.00 %[15:25:53]    + mixdown: Dolby Pro Logic II[15:25:53]    + encoder: faac
[15:25:53]      + bitrate: 192 kbps, samplerate: 48000 Hz
[15:25:53] reader: first SCR 0 id 224 DTS 31443
[15:25:53] yadif thread started for segment 1
[15:25:53] decomb thread started for segment 1
Encoding: task 1 of 1, 0.00 %
[15:25:53] decomb thread started for segment 0
[15:25:53] yadif thread started for segment 0
[15:25:53] mpeg2: "" (1) at frame 0 time 3003
[15:25:53] encx264: min-keyint: auto (29), keyint: 300
[15:25:54] encx264: encoding with stored aspect 1/1
[15:25:54] encx264: Encoding at constant RF 22.000000
Encoding: task 1 of 1, 0.00 %
x264 [info]: using SAR=1/1
x264 [info]: using cpu capabilities: MMX2 SSE2Fast SSSE3 FastShuffle SSE4.1 Cache64
x264 [info]: profile High, level 4.0No accelerated IMDCT transform found
...
Encoding: task 1 of 1, 100.00 %
[00:02:44] work: average encoding speed for job is 3.914453 fps
[00:02:45] sync: got 121393 frames, 120980 expected
Encoding: task 1 of 1, 100.00 %
[00:02:45] mpeg2 done: 121394 frames
[00:02:45] render: lost time: 0 (0 frames)
[00:02:45] render: gained time: 0 (0 frames) (0 not accounted for)
x264 [info]: frame I:508   Avg QP:19.91  size:145498  PSNR Mean Y:45.84 U:48.95 V:50.58 Avg:46.75 Global:46.37
x264 [info]: frame P:44296 Avg QP:22.93  size: 63789  PSNR Mean Y:43.01 U:46.31 V:48.39 Avg:43.94 Global:43.53
Encoding: task 1 of 1, 100.00 %
x264 [info]: frame B:76588 Avg QP:25.50  size: 21437  PSNR Mean Y:41.85 U:45.50 V:47.94 Avg:42.84 Global:42.44
x264 [info]: consecutive B-frames:  2.3% 31.3% 25.0% 41.5%
x264 [info]: mb I  I16..4: 13.3% 76.2% 10.5%
Encoding: task 1 of 1, 100.00 %
x264 [info]: mb P  I16..4:  3.3%  7.5%  0.4%  P16..4: 51.1% 18.7% 10.2%  0.0%  0.0%    skip: 8.8%
x264 [info]: mb B  I16..4:  0.4%  0.8%  0.0%  B16..8: 44.6%  5.4%  1.1%  direct: 7.1%  skip:40.5%  L0:43.7% L1:46.1% BI:10.2%
x264 [info]: 8x8 transform intra:67.3% inter:83.4%
...
x264 [info]: SSIM Mean Y:0.9808523 (17.179db)
x264 [info]: PSNR Mean Y:42.290 U:45.808 V:48.115 Avg:43.260 Global:42.817 kb/s:8969.49
[00:02:46] mux: track 0, 121392 frames, 4541319007 bytes, 8969.30 kbps, fifo 32
[00:02:46] mux: track 1, 189869 frames, 83597324 bytes, 165.11 kbps, fifo 256
[00:02:46] decomb: deinterlaced 67534 | blended 23456 | unfiltered 30402 | total 121392
Muxing: this may take awhile...[00:02:46] libhb: work result = 0

Rip done!
HandBrake has exited.
Old file size = 6891935748
New file size = 4628649970

[JohnAStebbins]

You have enabled both decomb and deinterlace filters. This won't cause the issue you see, but it will degrade picture quality. Use one or the other, not both.

Have you tried the detelecine filter? If the show is a movie that was originally shot at 24 fps, you would need this. It is generally safe to leave this filter on since it is only applied when it detects that it is needed.

[Deleted User 11865]

In the future, please redirect stderr and stdout to different files, and post only the stderr output. The progress (written to stdout) is not useful and makes the log less readable.

[luv2hike]

I'm not sure I know how to send stderr & stdout to different files. I just copy/pasted the output of the run from the terminal.

Can someone explain the difference between decomb, deinterlace, and detelecine? So the program detects if it needs detelecine so I can leave it, but it doesn't not detect interlacing or combing? Is one better than the other? My script has to be general purpose enough to handle whatever is thrown at it from another process so I cannot change it per video file. I just know I don't want the transcoded output to be interlaced but cannot guarantee that the input video is or isn't.

[luv2hike]

Oh and my wife finally noticed this issue with her glasses instead of contacts. It is like a very very fast strobe light so any fast head or hand motion is not exactly smooth but kind of stuttering, though very subtly and almost imperceptible. The key word is almost. The fact we can notice it tells me there is something that needs tweaking. I've tried a range of -q values and it doesn't affect this issue. Could it be the deinterlace/decomb combo mentioned above?

[JohnAStebbins]

Can someone explain the difference between decomb, deinterlace, and detelecine?

Detelecine https://trac.handbrake.fr/wiki/Telecine
Deinterlace https://trac.handbrake.fr/wiki/DeinterlacingGuide
Decomb https://trac.handbrake.fr/wiki/Decomb

All from the HandBrake Guide https://trac.handbrake.fr/wiki/HandBrakeGuide

Could it be the deinterlace/decomb combo mentioned above?

You have enabled both decomb and deinterlace filters. This won't cause the issue you see...
...

Telecined video can cause an effect similar to what you describe. That's why I suggested trying the detelecine filter.

[luv2hike]

Thanks to all for the much appreciated help! I have now read through the guide on all this stuff (sorry, didn't know it existed before as I only found HandBrake this past week and missed the link for it), and it was useful but also information overload. However, after reading it and from the previous replies in this thread, I think I do not need the deinterlace filter but should use decombing, remove deblocking, and add detelecine. I am making this change and re-running. I'll post results once I see them.

I also read about the extra encoding options for x264. Now my head really hurts. For now, I'm leaving well enough alone on these unless someone can point out reasons to add them. Feeling a bit overwhelmed.

[luv2hike]

Okay, I ran a small sample of the HD video that seems jittery from the other thread (stopped the run after a minute of video had been encoded just to test the settings as I see the problem usually within the first 30 seconds) using this new command:

Code: Select all

HandBrakeCLI -i "$1" -o "$2" -f mkv -e x264 -q 23 -E faac -B 192 -R 48 --strict-anamorphic --detelecine --decomb

and I think the problem is still there, though maybe improved somewhat. Honestly, my eyes may be deceiving me now from over-staring at all these. If I put this 55MB clip on my server along with the same time saved out of the original (can "ffmpeg -vcodec copy -acodec copy -t 00:01:00" do this?), would it be helpful to see if anyone else notices what I think I'm seeing?
Here is the output from this last run:

Code: Select all

[11:51:47] file is MPEG DVD Program Stream
[11:51:48] add_audio_to_title: added AC3 audio stream 0x80bd
[11:51:48] scan: decoding previews for title 1
[11:51:48] scan: audio 0x80bd: AC-3, rate=48000Hz, bitrate=448000 Unknown (AC3) (5.1 ch)
Scanning title 1...
Scanning title 1...
[11:51:48] scan: 10 previews, 1920x1088, 29.970 fps, autocrop = 0/8/0/0, aspect 1.76:1, PAR 1:1
[11:51:48] scan: title (0) job->width:1920, job->height:1072
[11:51:49] libhb: scan thread found 1 valid title(s)
+ title 1:
  + stream: 5081_20110110185900-ORIG.mpg
  + duration: 01:07:15
  + size: 1920x1088, pixel aspect: 1/1, display aspect: 1.76, 29.970 fps
  + autocrop: 0/8/0/0
  + chapters:
    + 1: cells 0->0, 0 blocks, duration 01:07:15
  + audio tracks:
    + 1, Unknown (AC3) (5.1 ch) (iso639-2: und), 48000Hz, 448000bps
  + subtitle tracks:
    + 1, Closed Captions (iso639-2: und) (Text)(CC)
[11:51:49] 1 job(s) to process
[11:51:49] starting job
[11:51:49] sync: expecting 120980 video frames
[11:51:49] job configuration:
[11:51:49]  * source
[11:51:49]    + 5081_20110110185900-ORIG.mpg
[11:51:49]    + title 1, chapter(s) 1 to 1
[11:51:49]  * destination
[11:51:49]    + 5081_20110110185900-q23-CT.mkv
[11:51:49]    + container: Matroska (.mkv)
[11:51:49]  * video track
[11:51:49]    + decoder: mpeg2
[11:51:49]      + bitrate 20000 kbps
[11:51:49]    + frame rate: same as source (around 29.970 fps)
[11:51:49]    + strict anamorphic
[11:51:49]      + storage dimensions: 1920 * 1088 -> 1920 * 1080, crop 0/8/0/0, mod 0
[11:51:49]      + pixel aspect ratio: 1 / 1
[11:51:49]      + display dimensions: 1920 * 1080
[11:51:49]    + filters
[11:51:49]      + Detelecine (pullup) (default settings)
[11:51:49]      + Decomb (default settings)
[11:51:49]    + encoder: x264
[11:51:49]      + quality: 23.00 (RF)
[11:51:49]  * audio track 0
[11:51:49]    + decoder: Unknown (AC3) (5.1 ch) (track 1, id 80bd)
[11:51:49]      + bitrate: 448 kbps, samplerate: 48000 Hz
[11:51:49]    + mixdown: Dolby Pro Logic II
[11:51:49]    + encoder: faac
[11:51:49]      + bitrate: 192 kbps, samplerate: 48000 Hz
[11:51:49] reader: first SCR 0 id 224 DTS 31443
[11:51:49] yadif thread started for segment 1
[11:51:49] yadif thread started for segment 0
[11:51:49] decomb thread started for segment 0
[11:51:49] encx264: min-keyint: auto (29), keyint: 300
[11:51:49] encx264: encoding with stored aspect 1/1
[11:51:49] encx264: Encoding at constant RF 23.000000
x264 [info]: using SAR=1/1
[11:51:49] decomb thread started for segment 1
x264 [info]: using cpu capabilities: MMX2 SSE2Fast SSSE3 FastShuffle SSE4.1 Cache64
[11:51:49] mpeg2: "" (1) at frame 0 time 3003
x264 [info]: profile High, level 4.0
No accelerated IMDCT transform found
[11:59:39] reader: done. 0 scr changes
[11:59:39] work: average encoding speed for job is 4.734588 fps
[11:59:39] sync: got 2224 frames, 120980 expected
[11:59:39] mpeg2 done: 2242 frames
[11:59:39] render: lost time: 9009 (3 frames)
[11:59:39] render: gained time: 9009 (12 frames) (0 not accounted for)
[11:59:39] render: average dropped frame duration: 3003
x264 [info]: frame I:8     Avg QP:19.60  size:160013  PSNR Mean Y:46.04 U:51.33 V:51.91 Avg:47.23 Global:47.15
x264 [info]: frame P:629   Avg QP:22.48  size: 57604  PSNR Mean Y:42.54 U:47.90 V:48.74 Avg:43.75 Global:43.63
x264 [info]: frame B:1501  Avg QP:26.35  size: 11864  PSNR Mean Y:40.53 U:47.63 V:48.49 Avg:41.91 Global:41.74
x264 [info]: consecutive B-frames:  0.7%  3.4% 38.3% 57.7%
x264 [info]: mb I  I16..4: 14.0% 79.3%  6.7%
x264 [info]: mb P  I16..4:  2.5%  4.6%  0.3%  P16..4: 45.0% 16.8% 11.4%  0.0%  0.0%    skip:19.5%
x264 [info]: mb B  I16..4:  0.3%  0.4%  0.0%  B16..8: 36.9%  3.0%  0.6%  direct: 2.2%  skip:56.6%  L0:43.8% L1:49.8% BI: 6.3%
x264 [info]: 8x8 transform intra:63.7% inter:83.0%
x264 [info]: coded y,uvDC,uvAC intra: 44.1% 56.4% 10.7% inter: 17.1% 13.0% 0.4%
x264 [info]: i16 v,h,dc,p: 40% 28% 15% 18%
x264 [info]: i8 v,h,dc,ddl,ddr,vr,hd,vl,hu: 22% 18% 38%  3%  3%  4%  4%  4%  5%
x264 [info]: i4 v,h,dc,ddl,ddr,vr,hd,vl,hu: 26% 26% 16%  4%  6%  5%  7%  4%  6%
x264 [info]: i8c dc,h,v,p: 47% 25% 23%  5%
x264 [info]: Weighted P-Frames: Y:1.7% UV:1.7%
x264 [info]: ref P L0: 48.3% 10.7% 30.8% 10.1%  0.0%
x264 [info]: ref B L0: 69.0% 25.9%  5.1%
x264 [info]: ref B L1: 86.7% 13.3%
x264 [info]: SSIM Mean Y:0.9797239 (16.930db)
x264 [info]: PSNR Mean Y:41.145 U:47.723 V:48.579 Avg:42.474 Global:42.229 kb/s:6195.01
[11:59:39] mux: track 0, 2136 frames, 55292383 bytes, 6189.04 kbps, fifo 8
[11:59:39] mux: track 1, 3349 frames, 1439845 bytes, 161.17 kbps, fifo 256
[11:59:39] decomb: deinterlaced 1172 | blended 618 | unfiltered 414 | total 2204
HandBrake has exited.

EDIT: appended the rest of the log that I missed copying earlier.

[JohnAStebbins]

It looks like you truncated the log. There is usually a summary at the end of what frames decomb was applied to and whether detelecine was applied to any frame.

[JohnAStebbins]

From the looks of it, the detelecine filter doesn't think that video is telecined. It only dropped 3 frames that it thought looked telecined. The video appears to be interlaced since decomb deinterlaced a signficant number of frames. I wonder if the video really is telecined and the detelecine filter is failing for some reason on this content. Can you tell us more about this video? Where did it come from? Is it a film transfer?

[luv2hike]

It is recorded from an HD HomeRun capture device via MythTV. This particular show was on our local PBS HD station in 1080 resolution (not sure if progressive or interlaced though). I've placed a couple of clips on my server, one of the original MPEG-2 file and the other of the MKV created by HandBrake if that helps. The link is at:

http://www.bradandsteph.com/hbtest/

They are both only about 1:11 long. The original is approx. 110MB and the transcoded one is 55MB. It is not a streaming server though, just plain old DSL so it may be a tad slow to download. After the short intro credit screen, I notice the peculiarity in the head motions of the man & woman standing and talking. I'm not sure I'd notice it in the H264 version if I didn't have the original to compare with. It just "feels" better, even when viewing them "blind" without knowing which my wife selected to play. I know that's vague and I'm sorry. Maybe the clips can help?

[Lostless]

Sounds like missing frames. PBS HD is 1080i I assume since 1080P is not broadcasted. 1080i (or 480i for that matter) contains 60FPS of information. Handbrake can not take interlaced material and deinterlace to 60P. It converts down to 30P giving a slight jitter compared to the original source as half the motion information is gone. 720P or 1080P on the the other hand works great as there 60 Full frames to work with.
www.100FPS.com
good reference regarding this

[JohnAStebbins]

The samples look pretty normal to me. You could be seeing what Lostless is describing. It is very subtle. 1080i is 60 fields per second. There are 2 fields per frame and there are 30 frames per second. So each field is offset in time by a small amount. When deinterlacing, 2 fields are combined into a single progressive frame and the resulting output is 30 frames per second. So there is some temporal information lost.

Another possibility is that your playback device has some small issue with the resulting file. What are you using for playback?

[luv2hike]

Wow, the whole 60i vs. 30p thing actually sounds like it makes sense with what I perceive to be seeing. My device is my mythfrontend system using VDPAU connected to an LCD HDTV. It handles everything I've thrown at it, including 1080i videos from our camcorder. According to top, I use about 3.5% CPU while playing 1080i courtesy the nVidia GPU.

So if HandBrake doesn't like 60i input which causes the lost info that I'm seeing, how can I "preprocess" the file to feed it into HandBrake in a friendlier manner? Or am I asking for something that is impossible technically?

Again, thanks SO very much for the help!

[JohnAStebbins]

There's a method of deinterlacing called bobbing that converts 60i to 60p. Someone was working on that for handbrake, but it's kind of stalled right now. There are other tools that can do it. avisynth does it I think.

[luv2hike]

I just found this site via Google talking about this subject:

http://forum.videohelp.com/threads/3038 ... 0p-not-30p

The guy claims mencoder's yadif deinterlacer can do it as well. I'm not sure. I also checked all my other HD transcodes that look great and just verified they are indeed all 720p broadcast recordings. This one file that has been giving me fits is the only 1080i HD broadcast I've tried using HandBrake on. So I'm at least glad I'm not crazy and really was seeing something.

Assuming mencoder can really do it, would it then make sense to use it to deinterlace first, then run HandBrake to H264 without any filtering or would I be losing lots of quality running through 2 programs?

[JohnAStebbins]

It's kind of a tough call. mencoder can encode to h.264, but mencoder is really only good at producing avi files. avi doesn't properly support b-frames. so you might not be able to get good results by just encoding with mencoder. I would try encoding with mencoder, and remuxing to mkv with mkvmerge (if that's the final container you need). If that doesn't turn out well, you could encode with mencoder, but set the quality obscenely high (to loose as little as possible in this stage), then re-encode with handbrake.

[luv2hike]

I tried one short sample with mencoder and it was a train wreck. Either it really doesn't handle it well, or I really don't know how to use it. My guess is the latter.

I may have just stumbled upon another solution from a different angle (maybe). I cannot tell if my eyes are now playing tricks so I need my wife to take another look and see if she sees a difference. But it just occurred to me that since I'm playing these within MythTV and its player has a playback menu choice for Video Scan of auto detect, progressive, interlaced (normal), or interlaced (reversed), that I might be able to fix the issue during playback instead of while encoding the file. By default this setting is Auto-Detect. I read somewhere else online that some videos from certain 1080i broadcasts confuse this setting as they may jump back & forth from progressive frames to interlaced ones.

So I played back one of the encoded files from an earlier HB run and then selected Interlaced (normal). Wow, I think it looked better. So then I changed it to Interlaced (reversed) and now I think it looks darn near perfect! Of course, it could be wishful thinking in my mind's eye so I need a 2nd opinion re: the WAF. But if so, then I can just go ahead and encode these like I was doing with HB to save space and preserve picture quality, then on the 1080i videos, override the video scan detection in mythfrontend to fix the jittery motion.

What do you think? Does this at all make sense from a technical point of view? If I go this route, should I encode the 1080i settings WITHOUT decombing or deinterlacing, or would it matter? How about detelecining? I'm not sure it relates at all to this issue, so leave it on? Here's hoping...

[JohnAStebbins]

If you are encoding with HandBrakes decomb setting on, you should be selecting progressive from mythtv's video scan menu. You would only use the interlaced option if you were not removing the interlacing prior to viewing.

I don't think that encoding without decomb and using mythtv's deinterlacer should look any different than using decomb and telling mythtv the video is progressive. But you'll have to look at it and see.

For this particular source, you don't need detelecine. The original source is 30fps interlaced. But if the original source is film that is shot at 24fps progressive and transfered to 30fps interlaced, then you need to detelecine it to recover the original framerate. If you don't, there will be subtle jerkiness similar to what you've been describing. There's no simple way to detect when you need it and when you don't automatically. So if you are going to be encoding a variety of source material with the same settings, you should probably leave it on.