Juddering in encoded HD video

[akkers]

tsmuxer says 'Cant detect stream type'.

tssniper says 'Directshow problem - Failure while trying to render Filtergraph'

[poisondeathray]

when you cut it in avidemux, use format mpeg-ts (not mpeg-ps), video=> copy, audio=> copy

ctrl+s , use .ts extension, e.g. video.ts

probably a useless exercise, I'm 99% certain that the original file is the problem, or at least that specific one.

You said other files were ok. So it might be only certain files were damaged or processed incorrectly

[poisondeathray]

Another way you can cut it, mentioned earlier is dgindex

load the original video, use the "[" and "]" markers to mark in, mark out a section , file=>save project and demux video, then upload the .m2v elementary video here

[akkers]

Here is a cut using dgindex:
csrs.m2v

Using avidemux as instructed:
cars.ts

[poisondeathray]

That's confirms it. I would ask for recaptured files. Ask them to use hdvsplit

[akkers]

I would say that maybe it is a file/cature problem. Howevcer I also had the same problem with file1 (the guy with white shirt moving forward posted at start of this thread). That file1 caused me no end of headache and had same problem. However, after 2-3 weeks of banging my head its ok now. File1 can be exported ok no matter which settings I use. I just cannot get it to go wrong. I did nothing to that file.

So why did File1 misbehave and then started to be ok? This all degfies logic.

Can we establish that the cut files that I posted here today do actually show a problem or whether it is CS4 export.

[poisondeathray]

I would say that maybe it is a file/cature problem. Howevcer I also had the same problem with file1 (the guy with white shirt moving forward posted at start of this thread). That file1 caused me no end of headache and had same problem. However, after 2-3 weeks of banging my head its ok now. File1 can be exported ok no matter which settings I use. I just cannot get it to go wrong. I did nothing to that file.

So why did File1 misbehave and then started to be ok? This all degfies logic.

Can we establish that the cut files that I posted here today do actually show a problem or whether it is CS4 export.

Some of your files might be ok. If you recall, "walcut3.avi" was fine, fields were intact. It was true 50i. The only difference was AVI wrapper

Was that part of the same video as the car video ? or different clip

Without a doubt, that dgindex clip proves problem is in original file of the cars video (at least that clip or section of that clip)

[akkers]

I have cut that clip again using dgindex from file1 (walcut3). This file gave me headches for 2 weeks and then eventually started working ok. I did nothing to file1. I cant understand the logic.

file1ex.m2v


Could some parts of the files be ok and rest not ok and throwing the encoder/decoders off depending where the edit/encoding started?

When I first exported file2 (the one with young men and cars) the export file looked ok for the first 5 or so minutes. But then started to judder. The clip of the men was about 20mins into file2 and the cars were about 50mins into file2). Why was the first 5 mins not juddering? How can that be?

If I fail to have the footage captured again, what are the options? Can we repair it using a tool?

[File1 and file2 were from the sam cam of the same event on the same day.]

[poisondeathray]

I have cut that clip again using dgindex from file1 (walcut3). This file gave me headches for 2 weeks and then eventually started working ok. I did nothing to file1. I cant understand the logic.


[Attachment 10048 - Click to enlarge]

As before, this clip is ok too, there are a couple of misplaced fields in the beginning but it might be from the way it was cut

Could some parts of the files be ok and rest not ok and throwing the encoder/decoders off depending where the edit/encoding started?

When I first exported file2 (the one with young men and cars) the export file looked ok for the first 5 or so minutes. But then started to judder. The clip of the men was about 20mins into file2 and the cars were about 50mins into file2). Why was the first 5 mins not juddering? How can that be?

It's unlikely , but anything is possible I suppose

It depends on how the clip was processed. For example if the other party appended clips, some good, some bad, it might have sections that are ok interspersed with sections that are bad

I've never seen hardware produce results like that;. i.e. if you capture straight from the camera from the same shooting session, using the same camera settings, it should all be the same - not sections of 25p and 50i content . And the problem just isn't 25p vs. 50i content, there is a field shift as well, as jagabo described

These results suggest some human error involved. The fact that these were given to you as "mpg" and not "m2t" already suggests something has been done to them (not the original capture)

If I fail to have the footage captured again, what are the options? Can we repair it using a tool?

[File1 and file2 were from the sam cam of the same event on the same day.]

Re-capture properly is clearly the best option

You can attempt to interpolate missing fields using avisynth, but unless you are already familar with it, it will be quite a big learning curve. Also there are down sides and side effects, as mentioned earlier - it's not perfect.

[jagabo]

Get a $10 firewire card and transfer the files from the camcorder to the computer yourself.

[akkers]

I wonder if they had dirty heads when they played the tapes and did the capture. This could have produced footage of varying quality.

[jagabo]

I wonder if they had dirty heads when they played the tapes and did the capture. This could have produced footage of varying quality.

No.

Does your camcorder have both 30i and 30p modes? Maybe you shot some segments at one setting, others at the other.

Note that I continue to use the term 25i whereas edDV has started using 50i. They mean the same thing.

[akkers]

I wonder if they had dirty heads when they played the tapes and did the capture. This could have produced footage of varying quality.

No.

Does your camcorder have both 30i and 30p modes? Maybe you shot some segments at one setting, others at the other.

Note that I continue to use the term 25i whereas edDV has started using 50i. They mean the same thing.

The footage was done bya 3rd party using a semi-pro PAL cam - not sure which one. I guess they will have shot it in 50i.

[jagabo]

Here's wcars2 after phase alignment and motion interpolation to 50p with this AviSynth script:

Import("C:\Program Files (x86)\AviSynth 2.5\plugins\DoubleFPS2.avs")

Mpeg2Source("wcars2.d2v", CPU=2) 
AssumeTFF()
SeparateFields()
Trim(1,0)
Weave()
DoubleFPS2()

Of course, you would be better off getting the original 25i video.

[akkers]

Sorry for asking a silly question but is it possibel that there may be corrupt portions every few minutes which are throwing off th rest of correct footage?

I tried to download mpeg2repair but my viruscan software wont let me install it.

[aedipuss]

not in HDV video. the capture from tape to computer file is digital and since it's long gop mpeg-2 the dropping or loss of a single packet/frame will cause blank/black video of up to 1/2 a second or a complete dead spot. after the dropout the file would once again be perfect.

[akkers]

Bingo!!! I have today found some common sense to what has been happening during my adventure with these video footage files.

It appears that file2 inherited some problem during the capture stage and is indeed dodgy. The 3rd party supplier has accepted responsibility and I should get another copy.

However, file1 looked ok (walcut3 with guy in white shirt) and was ok all along. It appears that I was taking the original file1 mpeg2 file, editing in premiere and exporting to avch/H264 (or using tmpgenc to convert to avch/h264) to view on the tv. It now looks like this process was the root of all my problems. Just by accident today I edited file1 mpeg2 file in premiere and exported it as mpeg2 file. I took this file and played it on the tv and it played spot-on. Bingo.

So for file1 it was the conversion from mpeg2 to avch/h264 that was causing the problem. Can I concluded that:

(1) Conversion of mpeg2 to avch/h264 is doomed

or

(2) I have some issue with the mpeg2 or avch/h264 encoder on my pc

(3) mpeg2 to mepg2 is spot-on


I will do more tests tomorrow and report back.

[jagabo]

(1) Conversion of mpeg2 to avch/h264 is doomed

No. You can encode h.264 interlaced. Or you can downsize to 1280x720 and bob deinterlaced to 50 fps progressive.

[akkers]

Just an after-thought - when I work with well behaved interlace video clips I usually see interlace lines in virtualdub.

However, these clips file1 and file2 go through virtualdub and show no sign of interlace lines or combing. I am beginiing to wonder if this is indeed interlaced footage, even though its properties indicate interlaced.

[jagabo]

As was pointed out several times: the "bad" clips are encoded interlaced but contain progressive frames, with fields out of phase. I saw lots of combing when I opened them in VirtualDub.

[akkers]

A very big thanks to all members here for all the help. Its been a very important learning curve for me.

Thanks again.

[akkers]

A brief update on this dreaded saga:

File1 - someone on another forum suggested that editing in mpeg2 was not a good idea. He suggested that I should encode file1 to DNxHD and edit that. That I did and it worked. I have been able to edit and export file1 ok. Job done.

File2 was corrupt in some way; possibly through a flawed capture workflow. File2 was shot on a Sony HDV tape cam, which I believe, only shoots 1080i (no 25p option). I have asked the footage to be captured again.

[edDV]

File2 was corrupt in some way; possibly through a flawed capture workflow. File2 was shot on a Sony HDV tape cam, which I believe, only shoots 1080i (no 25p option). I have asked the footage to be captured again.

Depends on the camcorder. Remind me of the HDV cam model that did the shoot. While 1440x1080 25/29.97i are the normal default, the following also exist.

1440x1080 25/23.976pf* - PAL models use 2:2 interlace, NTSC models use 2:3 telecine.
1440x1080 25/23.976p - Both PAL and NTSC models use 2:2 interlace.
1440x1080 23.976 24pA* (advanced) - NTSC pro cams only. Uses a 2:3:3:2 telecine sequence

1280x720 25/29.97p - mostly a pro JVC variant
1280x720 50/59.94p* - while formally XDCAM HD/EX, some "HDV" pro cams also have this option.


* exists on some Sony models.

[jagabo]

File2 was corrupt in some way; possibly through a flawed capture workflow. File2 was shot on a Sony HDV tape cam, which I believe, only shoots 1080i (no 25p option). I have asked the footage to be captured again.

By File2 you mean wcars2.mpg in this post:

https://forum.videohelp.com/threads/341052-Juddering-in-encoded-HD-video?p=2126631&view...=1#post2126631

It looks like it was originally 25i interlaced video that was deinterlaced poorly to create 25p then a field delay added and encoded interlaced. So half the vertical information is gone. This can be seen as jagged near-horizontal edges:



That image was made by realigning the fields back to progressive frames with AviSynth:

Mpeg2Source("wcars2.d2v", CPU=2)
AssumeTFF()
SeparateFields()
Trim(1,0)
Weave()

You can do the same with VirtualDub's Field Delay filter.