[RESOLVED] Deblocking problem with DGIndexNV 2038

[BarryW]

Here's a 9 second (8.22MB) sample from BBC HD - it's the first 9 seconds of Mountain part 3.
http://www.mediafire.com/?ki7dd8uik1phzlp

When encoding with DGIndexNV 2038 you'll see a deblocking problem at 4 seconds when the scene change occurs.
When encoding with the old DGAVCIndexNV 1.0.13 you won't see the deblocking problem at 4 seconds.

The HD is from 2007 but it's recently been repeated on BBC HD.
I've seen this deblocking problem on a lot of stuff recently from BBC HD when using DGIndexNV 2038 to encode.
The problem usually occurs within 5 seconds from the start of a TS file.

The stats file (from x264) shows the input and output frame indexes are out-of-sequence.
I assume this is how the latest BBC HD encoder broadcasts the frames, i.e. out-of-sequence.
I've resorted to using DirectShowSource() in the AVS file along with ffdshow rev3814 to encode these files.
I'm using the latest x264 r1947 and megui 2008 (svn)

Just thought you'd be interested.

Since switching off the preview mode in megui the frame repeat problem at the end of an encode seems to have gone away.

[BarryW]

When running the AVS file from DGIndexNV 2038 through AvsP the scene change at 4 seconds occurs on frame 122.
When running the AVS file from DGAVCIndexNV 1.0.13 through AvsP the scene change occurs on frame 123.

[BarryW]

Here's some more information that might be of use.
I've not included the DGI and DGA files - you can re-create these from the TS file.

Excerpt from x264 stats file when using DGIndexNV 2038 (deblock problem occurs):
in:116 out:118 type:b dur:2 cpbdur:2 q:20.96 tex:33423 mv:10022 misc:3355 imb:80 pmb:1800 smb:1720 d:- ref:0 ;
in:118 out:119 type:b dur:2 cpbdur:2 q:20.90 tex:34822 mv:9150 misc:3332 imb:105 pmb:1677 smb:1818 d:- ref:0 ;
in:121 out:120 type:P dur:2 cpbdur:2 q:20.72 tex:62830 mv:13758 misc:2828 imb:369 pmb:2202 smb:1029 d:- ref:0 ;
in:120 out:121 type:b dur:2 cpbdur:2 q:20.79 tex:45137 mv:9879 misc:3040 imb:21 pmb:2123 smb:1456 d:- ref:0 ;
in:122 out:122 type:I dur:2 cpbdur:2 q:18.31 tex:1007877 mv:129542 misc:469 imb:3600 pmb:0 smb:0 d:- ref:; <---------------scene change
in:123 out:123 type:P dur:2 cpbdur:2 q:20.42 tex:853298 mv:29202 misc:412 imb:2014 pmb:1586 smb:0 d:- ref:0 w:6,57,7 ;
in:124 out:124 type:P dur:2 cpbdur:2 q:20.32 tex:829429 mv:29727 misc:716 imb:1449 pmb:2055 smb:96 d:- ref:0 ;
in:125 out:125 type:P dur:2 cpbdur:2 q:21.12 tex:808264 mv:31260 misc:484 imb:2024 pmb:1542 smb:34 d:- ref:0 ;

Excerpt from x264 stats file when using DGAVCIndexNV 1.0.13 (encodes ok):
in:118 out:118 type:B dur:2 cpbdur:2 q:21.26 tex:133529 mv:17963 misc:2244 imb:333 pmb:2571 smb:696 d:- ref:0 ;
in:117 out:119 type:b dur:2 cpbdur:2 q:21.29 tex:34507 mv:10089 misc:3236 imb:174 pmb:1530 smb:1896 d:- ref:0 ;
in:119 out:120 type:b dur:2 cpbdur:2 q:21.24 tex:30927 mv:10770 misc:3359 imb:250 pmb:1429 smb:1921 d:- ref:0 ;
in:122 out:121 type:P dur:2 cpbdur:2 q:21.11 tex:66290 mv:13115 misc:2883 imb:355 pmb:2013 smb:1232 d:- ref:0 ;
in:121 out:122 type:b dur:2 cpbdur:2 q:21.16 tex:29790 mv:9239 misc:3099 imb:28 pmb:1643 smb:1929 d:- ref:0 ;
in:123 out:123 type:I dur:2 cpbdur:2 q:18.62 tex:1484865 mv:131967 misc:488 imb:3600 pmb:0 smb:0 d:- ref:; <---------------scene change
in:126 out:124 type:P dur:2 cpbdur:2 q:20.82 tex:589746 mv:21301 misc:857 imb:744 pmb:2707 smb:149 d:- ref:0 ;

[admin]

Can you try with 2039 please? Download it fresh to be sure to have all the slipstreams I made.

2039 included a fix for BBC HD streams.

[BarryW]

Yes, I've just downloaded 2039 - I'll give a go in the morning.

[admin]

I tested your stream with 2038 and 2039. 2038 was broken but 2039 is fine. A fix was added for this specific problem.

I don't know what you are talking about when you talk about x264 stats and "out of sequence" frames. Are you identifying an issue with DGDecNV?

[BarryW]

Yes, 2039 works ok.
-----------
The stats file (created by x264 on the first pass) shows the 'in' and 'out' fields.
The 'in' and 'out' are specified at mewiki as the "input frame number" and "coded frame number".
I assume the out-of-sequence ordering of the input frame numbers was something the HD broadcast encoder does.
The frame numbers come into-sync on the 'I' frames (where a scene change or the start of another GOP occurs).

Excerpt from x264 stats file using DGIndexNV 2039 (scene change occurs on frame 122 the same as DGAVCIndexNV 1.0.13)
in:121 out:120 type:P dur:2 cpbdur:2 q:20.73 tex:74143 mv:13836 misc:2717 imb:353 pmb:2265 smb:982 d:- ref:0 ;
in:120 out:121 type:b dur:2 cpbdur:2 q:20.80 tex:43708 mv:10159 misc:3069 imb:36 pmb:2063 smb:1501 d:- ref:0 ;
in:122 out:122 type:I dur:2 cpbdur:2 q:18.32 tex:1516429 mv:133367 misc:508 imb:3600 pmb:0 smb:0 d:- ref:; <-------------- scene change
in:124 out:123 type:P dur:2 cpbdur:2 q:20.44 tex:606718 mv:19680 misc:458 imb:1077 pmb:2500 smb:23 d:- ref:0 ;
in:123 out:124 type:b dur:2 cpbdur:2 q:20.44 tex:110104 mv:16979 misc:3157 imb:165 pmb:2189 smb:1246 d:- ref:0 ;

[admin]

I asked if you were identifying an issu ewith DGDecNV and you did not affirm that, so I mark this resolved.

[BarryW]

The x264 stats file just tells you what DGIndexNV is feeding the x264 encoder. All I was identifying was the fact the the scene change (at 4 seconds) was occurring on a different frame number depending on which version of the tools were used - that struck me as a bit odd. I thought this might help you resolve the issue - that's all. 2039 seems to be working ok, so yes, the problem is resolved.

[admin]

DGAVCDec is not licensed for your use. All licenses have been withdrawn and it is not available. You cannot legally use it. I will not respond to any support requests involving it.

[BarryW]

ok, understood.