View Full Version : Random Error Blocks on frames - xvid bug?


Meat_PoPsiclez
10th May 2006, 02:49
The video in question was encoded with the intention of permanent archival, so high bitrate is being used (q 3.75 giving ~11 - 25mbit/s). At various moments (a recurring theme seems to be the last frame in a gvop, ie frame before keyframe) error blocks will appear. What is consistent is the blocks area on a frame. What isn't consistent is the blocks themselfs are different every decode (skip a frame forward then back and it's different every time). I'm posting some example images to outline the problem, and also a clip that shows it. The clip is appreciably huge, so please only dl it if you can assist. This clip is a single gvop and the frame in question is the last frame in the group. This gvop was pulled from the middle of an encode.

http://www.anti-logic.com/richard/xvidbug-1.png
http://www.anti-logic.com/richard/xvidbug-2.png
http://www.anti-logic.com/richard/xvid_error_chunk.avi

The errors persist (in the same incosistent manner) using ffdshow's vfw, koepi's dshow xvid binary, and even xbmc's ffdshow mplayer implement. What I really want to know is if it's an encoder issue, and the videos are damaged, or is it a decoding issue? I've already spent ~60 cpu hours transcoding video, it'd be a shame if it's all bunk :(

The errors do not occur when using similiar options with the mpeg matrix and bitrate based control, leading me to believe it's not a hardware problem.

What version & date of XviD are you using?

Koepi binary XviD-1.1.0-30122005 _Final Release_

What DShow filter are using to decode it?

Tested both dshow binary by Koepi (via wmp), and ffdshow's vfw interface (via vdub mod)

Does the bug occur with every encode you do or just the one clip?

Every encode with this setup

Your computer config. (i.e. Media Players, Hardware (specifically: graphics card, Operating System, CPU) )

Athlon Xp 2500+ 1GB pc2700, ATI Radeon 8500, Windows XP SP2

Please post all your options you using in your encode (i.e. 2pass, GMC, Q-Pel, Bitrate, VHQ, etc)

adaptive quantitizer, interlaced, q-pel, gmp, vhq (mode 4), motion precision 6, chroma motion, trellis quant, 300 frame i-frame max, eqm_v3ehr.xcm matrix, constant q 3.75, no b-frames

Please try to fit a very brief summary of your problem into the subject line (rather than just having a subject of "help!" or similar)

Please specify the container format you are encoding to/playing from.

AVI (adobe premiere pro 1.5 movie export)

shon3i
10th May 2006, 10:38
What you use to deinterlace video?

sysKin
10th May 2006, 15:39
What you use to deinterlace video?
It's interlaced isn't it O_O

[edit] 1000th post! and so useless........ :(

Sirber
10th May 2006, 19:38
1000th post! and so useless........ :(Congrats! ;) I'm about to go to 7000 :D

Meat_PoPsiclez
10th May 2006, 23:55
Yes the video is interlaced (and encoded as interlaced with xvid).

There are some other error blocks present due to the minidv tape transfer, please try to ignore those. I knew the tapes were in poor condition so I didn't immediately stop encoding when I saw the error blocks.

I'm beginning to think I might be hitting some buffer overflow in xvid, possibly in the motion handling, as it I've only noticed the errors in the scenes where the bitrate peaks (~>16mbit/s). In low motion scenes the bitrate trails off and there are no errors to be seen.

Meat_PoPsiclez
13th May 2006, 04:01
Just tried pushing the bitrate more by forcing it into intraframe only (max i frame interval 1)... hit 68mbit/s and produced a video that had severe corruption in the first frame and crashes the xvid decoder (vfw and dshow) after a handful of frames (maybe 30 frames)! Seems to me there is definitely an overflow occuring.

At q3.75 mpeg matrix 1 iframe max it's making 36mbit/s and produces error blocks on every frame. If anyone would like I can upload a sample of this too.

Can anyone verify this behaviour by making a suitably high bitrate encode (>20mbit/s at least)? If it's not just me, I think this deserves and entry into xvid's bug tracker.

Things I've tried:
forcing min and max quant to 2 and 31 to see if this is a trellis bug reminant. Success : no
disable trellis. Success: no (causes xvid decoder to crash after first frame!!)
disable gmc and qpel. Success: no
Mpeg Matrix. Success: no
Non Interlaced Source. Success: no.
Changed q to 5 (bitrate @32mbit/s). Success: no (less area of each frame contains artifacts)
Changed q to 15 (bitrate @21mbit/s). Success: no (less area of each frame contains artifacts, not all frames contain artifacts)
Changed q to 31 (bitrate @19mbit/s). Success: no (lessened again)

sysKin
13th May 2006, 05:11
If anyone would like I can upload a sample of this too.
Yes we definitely would like that.

Caroliano
14th May 2006, 04:20
Can anyone verify this behaviour by making a suitably high bitrate encode (>20mbit/s at least)? If it's not just me, I think this deserves and entry into xvid's bug tracker.
I made an short encode at 24Mbps with Xvid 1.2 and no error. Same setings than you, but it was progressive.

I can upload the sample if anyone wants...