DV transfer - timecode scene detection not working...

[suAve]

I'm transfering MiniDV from a JVC GR-D93. I have about 20 scenes on a 1hour tape.

When I used WinDV, Ulead VS9, or MovieMaker, I ended up with over 200 scenes being detected, some a few frames long, some several seconds. I made sure I was doing scene detection by start/stop and not optical.

I tried two different firewire cards, two different cables and two different computers. Also, tried this on about 5 tapes, with the same results. What's interesting is that the scenes detected are always the same.

I've been over this for some time now, and can't make any sense about it.

I found only one workaround so far that works (other than joining the scenes together) with the freeware version of scenalizer. After transfering the whole tape, scenalizer properly detects the scenes based on start/stop (ie datestamp).

What would be embedded in the DV stream that would cause this ? The timecode is continious and there are no dropped frames at all. I haven't have a chance to try this on another camera and googling so far only showed another user with the same camera having the same problems but with Scenalizer Live (ie paid version). It's like all the software are detecting something else in the stream and therefore detecting it as a change.

Any ideas ? Of course the camera is out of warranty...

TIA...

[redwudz]

If I understand you correctly, you just want the 20 scenes detected? I turn the detection off when using WinDV, to keep it from making false detections, but then you would have a continuous video file. It may be mistaking dark frames for a video scene change. I don't use the timecode for anything.

With WinDV, I usually just set Discontinuity threshold to 0 and Max AVI size to 1000000 and transfer it all and sort it out later. You may be able to adjust those two settings and get better detection, but it's not perfect. Recording a few seconds of video with the camera with the lens cap on between scenes helps to separate them.

This guide for WinDV may help: http://www.dvd-guides.com/content/view/33/59/

[suAve]

No, I don't want 'optical scene detection' but camera start/stop ie datestamp detection, sometimes also called timecode detection. That is, "press start, record video, then press stop" is what's consider a scene. Even if you don't record with the lens cap (I usually do a fade to black on stop), it wouldn't matter as the scene should be detected via timecode.

When you detect by scene via start/stop, the software is supposed to either mark it inside the DV stream captured, or have the option to split each scene in it's own file.

I do get a scene detected based on timestamp, but I also get 20+ more within the same scene. Even if it's a still scene, it detects a break. Most software allows you to 'join' these scenes, but it's a pain going thru 200+ per tape, when it should be only 20 or so.

Hope this clarifies...

[JohnnyMalaria]

Can you upload a short segment?

If you can capture (without scene detection) about 2 seconds worth of video that you know results in lots of clips, I can look at the embedded information in every frame and see if there is anything that is not complying with the DV specification.

It's also possible your heads are getting dirty and corrupting the information. The DV spec provides cunning ways to hide any visual errors but errors in the subdata (the timecode etc) can't be fixed. The DV spec calls for the timecode, date etc to be written multiple times within the data for one frame - if the scene detection software only looks for the first place and not all places, you will be more prone to error problems.

Feel free to PM me with a URL...

[suAve]

I'll try to do that later tonight (or tomorrow). Never thought about a head issue. The camera is about 2 1/2 years old, and only recorded about 11 tapes, and watched them about once each (ie approx. 22 hours of use).

I'm also trying to test out in a friends system, and test my tapes on his camera and his tapes on my camera and see what results we get (he has no problems at all).

Is there any free tool out there to view the subdata ? I'm curious now... Thanks in advance..

[JohnnyMalaria]

I'm not aware of free tool specifically for view the subdata, however our Enhanced DV Decoder will display the date, timecode, camera settings etc on each frame. You could use the trial version. Open a DV file in a suitable media player and step through each frame at a time and see if the date etc jumps all over the place....

[gadgetguy]

It is my understanding that this is exactly what WinDV does via the discontinuity threshold setting.
From the WinDV website:

To my knowledge it is not capable of detecting scenes based on content.

[suAve]

Once in a while you get lucky with Google...

http://www.geocities.com/edward_gelzin

A Free AVI Analyzer. Basically it spits out all encoded data on the stream. And in my case, it seems that once in a while the timestamp would skip a second, or too many frames would have the same second:

09.07.2004 11:04:40 Frame 568 - TimeCode 00:00:18 13
09.07.2004 11:04:42 Frame 598 - TimeCode 00:00:19 13
09.07.2004 11:04:43 Frame 658 - TimeCode 00:00:21 13
09.07.2004 11:04:44 Frame 688 - TimeCode 00:00:22 13
09.07.2004 11:04:46 Frame 718 - TimeCode 00:00:23 13
09.07.2004 11:04:48 Frame 778 - TimeCode 00:00:25 13

So WinDV set at the default of 1 second would detect this as different scenes. It's random it seems as to when it happens...

I guess Ulead has the same default of 1 sec difference and maybe that's what's triggering it ? I only scanned a few seconds from one of the tapes, several times, and always get the same results - I guess the data was recorded like this...

[gadgetguy]

Nice find. Looks like a useful tool. So based on the data you're receiving WinDV did as it was told. Have you tried increasing the threshold to 5 or 6 seconds and trying your capsfer again to see if you get the scenes split as you expected?

[suAve]

Yes - just going to two seconds did the trick... I guess the default of one second was too much.

Now, trying to figure out how to get ulead VideoStudio to decrease the sensitivity as I'd rather one file and work then with the individual scenes from within VS10, than dealing with separate files...

I seriously doubt I'll get anything from JVC - The camera is out of warranty and 2 years old. Also, the fact that I never formatted the tapes, they'll claim I had to do that (though it's bogus)...

[gadgetguy]

You can always set the discontinuity to Zero and capsfer as one big file (as long as your max filesize is set high enough). This is what I normally do.
Under normal conditions there isn't any need to 'format' a MiniDV tape, however if you leave unrecorded gaps on the tape between scenes it will play havoc with your capsfer software. This can happen if you play a tape to review what's on it, and stop the tape after seeing static (or blue screen or however your cam handles it), then start recording again. To avoid this, some people record to the entire tape with the lenscap on before recording the 'real' video.