Avisynth lost frames

[vanden]

Hello,

Here is my problem, I have videos (from a Sony camera) MTS container:
Format/Info : Advanced Video Codec
Format profile : High@L4.0
Format settings, CABAC : Yes
Format settings, ReFrames : 2 frames
Format settings, GOP : M=2, N=13
Codec ID : 27
Bit rate mode : Variable
Bit rate : 21.2 Mbps
Maximum bit rate : 22.0 Mbps
Width : 1 920 pixels
Height : 1 080 pixels
Frame rate : 25.000 fps
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Interlaced
Scan order : Top Field First

I use première CS6 to video editing.
As the PC I use is not very powerful I converted the H264 video in Huffyuv (ffdshow).
To do this conversion I use Avisynth.
The problem is : whether with DGAVCindex or DirectShowSource I lost frames !
2 at the beginning of video with DGAVCindex
3 at the end with direcshowsource …

what solution exist ?

[Guest]

Please post an unprocessed source sample including the start of the file. You can cut it with DGSplit.

[vanden]

OK, here the sample (http://www.infonetservices02.fr/00001.MTS)

if i demus this file with eac3to :
eac3to v3.27
command line: "C:\Program Files (x86)\MeGUI_2028_x86\tools\eac3to\eac3to.exe" "E:\00001.MTS" 1:"E:\T1_Video - .h264" 2:"E:\T2_Audio - .ac3" -progressnumbers
------------------------------------------------------------------------------
M2TS, 1 video track, 1 audio track, 1 subtitle track, 0:00:07, 50i
1: h264/AVC, 1080i50 (16:9)
2: AC3, 2.0 channels, 256kbps, 48kHz
3: Subtitle (PGS)
[v01] Extracting video track number 1...
[a02] Extracting audio track number 2...
[v01] Creating file "E:\T1_Video - .h264"...
[a02] Creating file "E:\T2_Audio - .ac3"...
Video track 1 contains 364 fields.
eac3to processing took 1 second.
Done.
364 fields = 182 frames

if i load avs script :
http://infonetservices02.free.fr/DGAVCindex.jpg

if i load the mts file in premiere :
http://infonetservices02.free.fr/premiere.jpg
7sec + 7 frames = 7*25 +7 = 175 +7 = 182 frames.

[vanden]

corrected links errors !

[Guest]

Please also give us your Avisynth script.

[poisondeathray]

As the PC I use is not very powerful I converted the H264 video in Huffyuv (ffdshow).

If this is the reason you're doing this, an alternative method is using ffmbc (you can batch convert if it suits your needs)

ffmbc -i input.mts -vcodec ffvhuff -acodec copy -sn output.avi

note: -vcodec huffyuv is 4:2:2 only and will upsample the chroma, you can use that if you want; ffvhuff variant allows 4:2:0

[vanden]

My Avisynth script (DGAVCindex):
LoadPlugin("C:\Program Files (x86)\MeGUI_2028_x86\tools\dgavcindex\DGAVCDecode.dll")
AVCSource("C:\Users\Erik\Desktop\test\00001.dga")
My .dga file (http://infonetservices02.fr/00001.dga)



My Avisynth script (Directshowsource) :
LoadPlugin("C:\Program Files (x86)\MeGUI_2028_x86\tools\avs\directshowsource.dll")
DirectShowSource("D:\PRIVATE\AVCHD\BDMV\STREAM\00001.MTS", fps=25.000, audio=false, convertfps=true).AssumeFPS(25,1)

[Guest]

The sample has an open GOP, so the first two frames are B frames. They are not decodable. So DG tools replace them with a copy of the first decodable frame, i.e., frame 2 (0-based). Other tools may throw them away. You aren't losing anything that is actually available in the stream. DirectShowSource() is not frame accurate so anything you get from that is going to be a crap-shoot.

[Guest]

Hi rean. Of course many camcorders need special handling for joining. But may I ask specifically how Windmotion deals with a leading open GOP as we have here? Thank you.

[vanden]

You aren't losing anything that is actually available in the stream.

Here the 3 first frames (mts decoded by premiere) :
Frame0 (http://infonetservices02.free.fr/Frame-00.jpg)
Frame1 (http://infonetservices02.free.fr/Frame-01.jpg)
Frame2 (http://infonetservices02.free.fr/Frame-00.jpg)

Her the first frames of my Avisynth script (DGAVCindex) encoded in Huffyuv (ffdshow) and decoded by premiere :
Frame0 (infonetservices02.free.fr/Frame-00-DGAVCindex.jpg)

I have lost 2 frames.

[Guest]

Of course the program cannot recreate not existing frames from another MTS file. Then what do you do, throw away those frames? Or make up something sensible and include them?

To clarify, I am talking about when your first MTS has an open GOP.

[Guest]

It's possible the GOP is wrongly marked open. Investigating...

[poisondeathray]

If you are able to feed your same sample via Avisynth to some other application that sees two correct leading frames (not copies), please tell me about it.



ffmbc decodes this sample as 364 unique fields , or 182 unique frames (when examining the lossless intermediate) , same as premiere - no frames are lost

ffmbc handles interlaced avc transport streams (e.g from some camcorders, sat boxes) that some programs, and source filters have problems with . I have no idea how it works with these problem files. The sources are available.

somebody compiled a ffms2 based on ffmbc (instead of ffmpeg) , but the special whatever-it-is didn't carry through

[Guest]

Looks like it may be a simple bug in DGDecodeNV. Standby...

[Guest]

OK, I have it fixed. I forgot in a strategic place that an IDR refreshes everything. :)

I'll link a test version of DGDecodeNV.dll in a while, but just 32-bit. Is that OK?

[Guest]

Try this:

http://neuron2.net/misc/DGDecodeNV_vanden.zip

I was planning to add an "Output Trimmed VOB" feature, so that together with this fix will make a release.

Please advise your test results on various files. Thank you.

You are out of luck with DGAVCDec. I no longer distribute or maintain it and the source code is not in the public domain.

[feisty2]

use "LWLibavsource".selecteven for interlaced h264 ts files

[vanden]

Try this:

http://neuron2.net/misc/DGDecodeNV_vanden.zip

I was planning to add an "Output Trimmed VOB" feature, so that together with this fix will make a release.

Please advise your test results on various files. Thank you.

You are out of luck with DGAVCDec. I no longer distribute or maintain it and the source code is not in the public domain.

I do not understand what to do.

it must use DGIndexNV ?
I do not have a CUDA card.

Edit :
I tested on another PC (with nvidia card) but I get an error :
http://infonetservices02.free.fr/dgindexnv.png

[vanden]

@ feisty2 : how to use "LWLibavsource" ?

@ poisondeathray : Ok I will test ffmbc

[vanden]

I tested ffmbc:
ffmbcx64 -i 00001.MTS -vcodec ffvhuff -acodec copy -sn 00001.avi

Indeed, no loss of frames, but the resulting avi file is not opening in premiere !


Edit :
After remux (with no sound) by virtualdub, file is opening in premiere

What does it take command line to exclude the audio ? (or just demux audio).
Is it possible to activate "adaptive huffman table" when encoding the video ?

[Guest]

I tested on another PC (with nvidia card) but I get an error You need a license that you can obtain at my website.

[poisondeathray]

What does it take command line to exclude the audio ? (or just demux audio).


-an instead of -acodec copy



Is it possible to activate "adaptive huffman table" when encoding the video ?

Not sure


Or perhaps ffvhuff variant already does
no, ffvhuff == huffyuv + "adaptive huffman tables"

[vanden]

-an instead of -acodec copy


Thank, it works perfectly.
Or perhaps ffvhuff variant already does
no
add "-context 1" active "adaptive huffman table"