Log in

View Full Version : bug in interlaced mode


Davesmitty
7th August 2010, 18:53
There seems to be a bug in the interlaced mode of the newer versions of x264(This encode was created with rev.1688. from the x264.nl site)

commandline: "D:\Program Files\MeGUI\tools\x264\x264.exe" --level 3.2 --pass 1 --bitrate 2000 --stats "E:\test\test.stats" --slow-firstpass --deblock -1:-1 --keyint 60 --min-keyint 2 --b-adapt 2 --b-pyramid none --interlaced --vbv-bufsize 12000 --vbv-maxrate 8000 --direct auto --no-dct-decimate --no-fast-pskip --no-psy --aud --nal-hrd vbr --psnr --ssim --sar 8:9 --output NUL "E:\test\test.avs"

Every 50-100 frames there will be a frame that is slightly deformed. I checked to see if this was a problem with avisynth, but there were no similar defects when encoded with HCenc. I only noticed the defects when looking at individual frames with low psnr scores in MSU VQMT tool. I didn't do an tests with 1080i to see if similar defects were produced.

Blue_MiSfit
8th August 2010, 01:02
Hi there,

A few suggestions

1) Don't host images here on doom9 via attachments. They take forever to get approved, and cost doom9 money :) I'd suggest using imgur or something free that can direct link into the forum
2) Screenshots show very little. Provide a small piece of the source, and the encoding settings that can replicate the issues you're describing.
3) Upload the source to a free filesharing service other than rapidshare! I'm partial to Mediafire...

Your settings seem rather unusual. Are you using an up-to-date development build of MeGUI? The "release" build is HORRIBLY out of date and doesn't play too nice with x264's "new" CLI syntax last I checked.

Nevertheless, I matched your settings using a 480i60 stream I had lying around and wasn't able to notice anything obviously broken "every 50-100 frames", as you say, using x264 build 1688.

You may be seeing something akin to keyframe pumping, since your keyint is ridiculously low for 2mbps 480i ;)

Forget about MeGUI for a moment, and try something on the command line:

x264.exe input.avs --crf 18 --interlaced --preset slow --tune film --output output.mp4


How does that look?

Also, welcome to doom9! I'd be willing to guess you've been lurking for awhile? Bravo!

Derek

Trahald
8th August 2010, 01:18
pictures approved but feel free to edit links to an image hosting site instead.

Blue_MiSfit
8th August 2010, 01:46
Hmm... after jiggling your images back and forward I do see the strange artifact you're describing. The boxer to camera right looks like half his body moves but the other doesn't. Are you positively sure that these screenshots are of the same exact frame?

We need a source sample that will reliably produce these types of artifacts when using your odd choice of settings :)

Davesmitty
9th August 2010, 19:22
I tried out your commandline in the X264 cli and no artifacts were produced. So I figured that the megui interface may be causing the problem, but I tried out my commandline in the X264 cli. The artifacts were reproduced again. I then went back to the defaults and started adding 1 different switch(from my commandline) per encode to see what is causing the problem. The problem occurred when MV prediction mode is set to auto(--direct auto) and temporal(--direct temporal). There were no artifacts in the spatial( default mode).

LoRd_MuldeR
9th August 2010, 20:36
"direct=temporal + interlaced" is supposed to work since r1430:
http://git.videolan.org/gitweb.cgi?p=x264.git;a=commit;h=58fd85453ec62d829663887b960ff7f2fbbc22d1

What you are experiencing might be a regression though...

Dark Shikari
9th August 2010, 21:02
"direct=temporal + interlaced" is supposed to work since r1430:
http://git.videolan.org/gitweb.cgi?p=x264.git;a=commit;h=58fd85453ec62d829663887b960ff7f2fbbc22d1

What you are experiencing might be a regression though...Or perhaps his decoder is old and buggy. I regularly do regression tests involving interlaced and direct temporal and none currently fail.

SeeMoreDigital
9th August 2010, 21:08
If anybody wants, I can test some short interlaced samples in my hardware player/decoder.

Dark Shikari
9th August 2010, 21:12
If anybody wants, I can test some short interlaced samples in my hardware player/decoder.It's probably easier to just use JM...