Interlaced Telcine Footage?

[elliot_123]

I have a Sony DCR-PC1000 and it says it records in movie mode which applies a 24 frame like recording to your video clips. It seems the be some sort of an interlaced-progressive mix of frames or something. The video jumps\skips on playback. But I analyzed the clip in avidemux and separated the fields. I noticed that when the motion stops in the top field it continues in the bottem field for one frame and than catches up in the top field and another 4 frames plays and this process repeats. I could never figure out how to get the video to play back smoothly, I tried Handbrake and waited like an hour for the convert and it dident seem to make any progress. I was wondering if anyone else knew anything about this?

HERE IS A SAMPLE OF THE VIDEO FROM THE DCR-PC1000(MOVIE MODE): Captured with WinDV

http://www.mediafire.com/?732ym2ml0gyqsb3

[jagabo]

It looks to me like the camera takes 30 images per second, throws out one of every five, applies 2:3:3:2 pulldown on the remaining four to create 60 fields per second (four frames become ten fields), then packages pairs of fields together to produce 30 interlaced BFF frames per second. It will never play smoothly because of the missing frame. This is the worst "24p" mode I've ever seen. In fact, I find it hard to believe Sony would even include such a mode! I'm adding it to my collection of odd video clips.

I was able to use AviSynth and MVTools to restore the missing frames with motion interpolated frames.

import("C:\Program Files (x86)\AviSynth 2.5\plugins\ReplaceFramesMC.avs")

AviSource("C:\Users\John\Desktop\dvt2.12-05-06_17-32.00.avi") 
SelectEvery(5,0,2,2,3,4)
ReplaceFramesMC(2,1)
ReplaceFramesMC(7,1)
ReplaceFramesMC(12,1)
ReplaceFramesMC(17,1)
ReplaceFramesMC(22,1)
ReplaceFramesMC(27,1)
ReplaceFramesMC(32,1)
ReplaceFramesMC(37,1)
...

This type of motion interpolation screws up on some types of material. In this video it has problems with the vertical poles but worked fairly well on the trees and other stuff.

[elliot_123]

Where do you get the ReplaceFramesMC.avs file, if I want to use this script on other videos recorded like this, by the way it did a very good job getting rid of the motion judder, that was impressive.

[DB83]

This may be a stupid comment but I did not think that DV had a 24P mode. If the camera has recorded in 24P, this 'capture' is now 30i which surely has created all these artifacts.

[jagabo]

Where do you get the ReplaceFramesMC.avs file

It was posted somewhere around here under a different name:

function ReplaceFramesMC(clip Source, int N, int X)
{
 # N is number of the 1st frame in Source that needs replacing. 
 # X is total number of frames to replace
 #e.g. RX(101, 5) would replace 101,102,103,104,105 , by using 100 and 106 as reference points for mflowfps interpolation
 
start=Source.trim(N-1,-1) #one good frame before, used for interpolation reference point
end=Source.trim(N+X,-1) #one good frame after, used for interpolation reference point
 
start+end
AssumeFPS(1) #temporarily FPS=1 to use mflowfps
  
super = MSuper()
backward_vec = MAnalyse(super, isb = true)
forward_vec = MAnalyse(super, isb = false)
MFlowFps(super, backward_vec, forward_vec, blend=false, num=X+1, den=1) #num=X+1
AssumeFPS(FrameRate(Source)) #return back to normal source framerate for joining
Trim(1, framecount-1) #trim ends, leaving replacement frames
  
Source.trim(0,-N) ++ last ++ Source.trim(N+X+1,0)
}

https://forum.videohelp.com/threads/322848-Scripts-and-filters-for-an-Opera-%28sample-v...=1#post2091091

This may be a stupid comment but I did not think that DV had a 24P mode. If the camera has recorded in 24P, this 'capture' is now 30i which surely has created all these artifacts.

Normally, DV camcorders that support 24p shoot at 24p then package as 30i with 2:3:3:2 pulldown. That looks slightly more juddery that normal 2:3 pulldown used for film sources. But this camera shoots 30p, decimates to 24p, the applies 2:3:3:2 pulldown. So not only does it have pulldown judder it has a jerk every 4 frames from that missing frame. Even if you remove the pulldown you still have that missing frame.

[DB83]

Ok. The point I was trying to make is that the 'capture' does not accurately IMHO reflect what is in the original recording.

Surely there has to be a way to transfer the footage from the camera to an editor without the possibility of bastardising it any more.

Or to put it another way, are we certain that the recording was packaged as 30i ?

[jagabo]

Ok. The point I was trying to make is that the 'capture' does not accurately IMHO reflect what is in the original recording.

He captured from a DV camcorder using WinDV. That should give an exact copy of what's on the DV tape. He didn't mention doing any processing, and the sample is DV AVI (what WinDV outputs), so I assume he didn't mess with the video.

Or to put it another way, are we certain that the recording was packaged as 30i ?

Download the sample and look for yourself. It's obvious.

[DB83]

I did download it hence my (probably) half-cocked remarks about the 24P mode.

I only see black and white (no grey). So I expect if something is stored as 24P it should remain as 24P. Here, if I read correctly, the camera shoots at 30i, internally coverts to 24P and when this is transfered out it comes back at 30i.

I therefore fail to see the sense in using a 24P internal mode if you can not use that outside the camera.

But I probably have this wrong so I will now retire to my cave

[jagabo]

As I said, the camera appears to shoot 30p, decimate to 24p, then apply pulldown to 30i. Yes, there's no sense in shooting 24p with this camcorder as it results in jerky video. If the OP wants progressive frames he would be much better off if the camcorder has a 30p in 30i mode.

[edDV]

This is what Camcorderinfo had to say about the PC1000 in "24p" mode.

Scan Rates/24P (5.0)
The DCR-PC1000 has the same fake 24P mode as the DCR-PC350. The fake 24P tries to give a cinematic 24P effect, but it really just produces jittery video. We also discovered a 30P mode which looks much better, but apparently there is some confusion on whether this is a supported feature or we discovered something we shouldn't have.

http://www.camcorderinfo.com/content/Sony-DCR-PC1000-Camcorder-Review.htm

[elliot_123]

I captured the video in the standerd 60i on the camera. I even went as far as opening Avidemux, went to Video menu, clicked filters and Interlacing and selected "Stack Fields". I analyzed the 2 seperate fields for motion differences and found something interesting. I have also posted a screenshot of this.



I frame advanced in the video:

Frame 1: Top field starts motion
Frame 2: Bottom Field Catches up to top field(by 1 frame)
Frame 3: Both Fields begin moving normolly(motion wise)
Frame 6: Top field is ahead of bottem field(while motion catches up on bottem field)
Frame 7: Bottem field catches up to top field(by 1 frame)
Frame 11: Top field is ahead of bottem field
Frame 12: Bottem field catches up to top field(by 1 frame)
Frame 16: Top Field is ahead of bottem field

This process seems to go on and on in the same pattern in the video, but it proved that the motion is in the video but in the other field, it seems to wait for the other field to catch up or something.

[jagabo]

Use a bob instead of stacking. It's much more obvious what's going on.

[elliot_123]

Here is what it looked like using DG BOB in avidemux. You can see the motion artifacts from the other field

[jagabo]

Those are problems caused by improper chroma handling. Ignore them. Just watch the repeated frames. If they appear out of order you picked the wrong field order. You'll see your video has a 2:3:3:2 repeat pattern after bobbing.

VirtualDub's works much better than AviDemux for DV AVI sources.

[Gavino]

Normally, DV camcorders that support 24p shoot at 24p then package as 30i with 2:3:3:2 pulldown. That looks slightly more juddery that normal 2:3 pulldown used for film sources.

Out of interest, what is the reason for using 2:3:3:2 instead of 2:3?

[edDV]

Normally, DV camcorders that support 24p shoot at 24p then package as 30i with 2:3:3:2 pulldown. That looks slightly more juddery that normal 2:3 pulldown used for film sources.

Out of interest, what is the reason for using 2:3:3:2 instead of 2:3?

2:3:3:2 is called 24pA (progressive advanced) explained here. Part 2-2.
http://en.wikipedia.org/wiki/24p#Advanced_pulldown

A better explanation here IMO by Adam Wilt.
http://www.adamwilt.com/24p/

I seriously doubt it was Sony's intention to make this consumer camcorder true 24pA. It is more likely a non-compatible version. Sony Vegas Pro will handle true 24pA.

[jagabo]

It's easier to restore the original film frames after 2:3:3:2 pulldown (compared to 2:3 pulldown). Both have the same overall result, 4 film frames become 10 video fields (5 video frames when fields are woven into frames). Ie, 24 film frames becomes 60 video fields (30 frames when fields are woven into frames). But the difference is:

2:3 pulldown:
four film frames: 1 2 3 4
after 2:3 pulldown: 1t 1b 2t 2b 2t 3b 3t 4b 4t 4b
woven into frames: 1t1b 2t2b 2t3b 3t4b 4t4b

So four film frames have become 5 video frames. Three of the video frames are progressive in the sense that they include both the top and bottom fields from the same film frame (1t1b 2t2b 4t4b). Two of the video video frames are interlaced because they contain fields from different film frames (2t3b 3t4b). To restore the original film frames you can use the progressive frames directly (1 2 4). But to get film frame 3 you have pull its fields from the two interlaced frames and recombine them.

2:3:3:2 pulldown:
four film frames: 1 2 3 4
after 2:3:3:2 pulldown: 1t 1b 2t 2b 2t 3b 3t 3b 4t 4b
woven into frames: 1t1b 2t2b 2t3b 3t3b 4t4b

Once again, four film frames has become 5 video frames. But this time four of the video frames are progressive and contain all four of the original film frames (1t1b 2t2b 3t3b 4t4b). One of the video frames looks interlaced and contains fields from two adjacent film frames (2t3b). To restore the original film frames you just throw out that interlaced frame.

[Gavino]

Thanks to both edDV and jagabo for your comprehensive replies.

[elliot_123]

It's easier to restore the original film frames after 2:3:3:2 pulldown (compared to 2:3 pulldown). Both have the same overall result, 4 film frames become 10 video fields (5 video frames when fields are woven into frames). Ie, 24 film frames becomes 60 video fields (30 frames when fields are woven into frames). But the difference is:

2:3 pulldown:
four film frames: 1 2 3 4
after 2:3 pulldown: 1t 1b 2t 2b 2t 3b 3t 4b 4t 4b
woven into frames: 1t1b 2t2b 2t3b 3t4b 4t4b

So four film frames have become 5 video frames. Three of the video frames are progressive in the sense that they include both the top and bottom fields from the same film frame (1t1b 2t2b 4t4b). Two of the video video frames are interlaced because they contain fields from different film frames (2t3b 3t4b). To restore the original film frames you can use the progressive frames directly (1 2 4). But to get film frame 3 you have pull its fields from the two interlaced frames and recombine them.

2:3:3:2 pulldown:
four film frames: 1 2 3 4
after 2:3:3:2 pulldown: 1t 1b 2t 2b 2t 3b 3t 3b 4t 4b
woven into frames: 1t1b 2t2b 2t3b 3t3b 4t4b

Once again, four film frames has become 5 video frames. But this time four of the video frames are progressive and contain all four of the original film frames (1t1b 2t2b 3t3b 4t4b). One of the video frames looks interlaced and contains fields from two adjacent film frames (2t3b). To restore the original film frames you just throw out that interlaced frame.

Is their some sort of a script somebody has or can write for avisynth or something so that I can restore the film frames from the 60i material. In other words the 2 frames that are interlaced, can the full film frames be restored from those fields, according to what I read from the post above.

[jagabo]

Is their some sort of a script somebody has or can write for avisynth or something so that I can restore the film frames from the 60i material. In other words the 2 frames that are interlaced, can the full film frames be restored from those fields, according to what I read from the post above.

There are many IVTC filters in AviSynth. TFM().TDecimate() is one of the most reliable:

WhateverSource("fileName.ext")
AssumeTFF() #or AssumeBFF()
TFM()
TDecimate()

[elliot_123]

[QUOTE=elliot_123;2164054]

It's easier to restore the original film frames after 2:3:3:2 pulldown (compared to 2:3 pulldown). Both have the same overall result, 4 film frames become 10 video fields (5 video frames when fields are woven into frames). Ie, 24 film frames becomes 60 video fields (30 frames when fields are woven into frames). But the difference is:

2:3 pulldown:
four film frames: 1 2 3 4
after 2:3 pulldown: 1t 1b 2t 2b 2t 3b 3t 4b 4t 4b
woven into frames: 1t1b 2t2b 2t3b 3t4b 4t4b

So four film frames have become 5 video frames. Three of the video frames are progressive in the sense that they include both the top and bottom fields from the same film frame (1t1b 2t2b 4t4b). Two of the video video frames are interlaced because they contain fields from different film frames (2t3b 3t4b). To restore the original film frames you can use the progressive frames directly (1 2 4). But to get film frame 3 you have pull its fields from the two interlaced frames and recombine them.

2:3:3:2 pulldown:
four film frames: 1 2 3 4
after 2:3:3:2 pulldown: 1t 1b 2t 2b 2t 3b 3t 3b 4t 4b
woven into frames: 1t1b 2t2b 2t3b 3t3b 4t4b

Once again, four film frames has become 5 video frames. But this time four of the video frames are progressive and contain all four of the original film frames (1t1b 2t2b 3t3b 4t4b). One of the video frames looks interlaced and contains fields from two adjacent film frames (2t3b). To restore the original film frames you just throw out that interlaced frame.

But how would I restore film frame 3 by recombining the fields. Is their some sort of script or something besides decimate because I tried something like that and it was jerky for like 1 frame out of every 5.

[jagabo]

But how would I restore film frame 3 by recombining the fields. Is their some sort of script or something besides decimate because I tried something like that and it was jerky for like 1 frame out of every 5.

IVTC isn't the problem with your clip. The problem is your camcorder shot 30p, threw away one frame to make 24p, then applied 2:3:3:2 pulldown. You can restore the 24p video but you can't restore the frame the camera discarded before the pulldown. It doesn't exist any more. That's why I used ReplaceFramesMC() -- to generate that missing frame.

So your camera shot 30p: 1 2 3 4 5
It discarded one frame to make 24p: 1 2 4 5
Then performed 2:3:3:2 pulldown: 1t 1b 2t 2b 2t 4b 4t 4b 5t 5b
Packed as frames: 1t1b 2t2b 2t4b 4t4b 5t5b

So you can restore frames 1 2 4 and 5. But frame 3 doesn't exist in the video.

[elliot_123]

But how would I restore film frame 3 by recombining the fields. Is their some sort of script or something besides decimate because I tried something like that and it was jerky for like 1 frame out of every 5.

IVTC isn't the problem with your clip. The problem is your camcorder shot 30p, threw away one frame to make 24p, then applied 2:3:3:2 pulldown. You can restore the 24p video but you can't restore the frame the camera discarded before the pulldown. It doesn't exist any more. That's why I used ReplaceFramesMC() -- to generate that missing frame.

So your camera shot 30p: 1 2 3 4 5
It discarded one frame to make 24p: 1 2 4 5
Then performed 2:3:3:2 pulldown: 1t 1b 2t 2b 2t 4b 4t 4b 5t 5b
Packed as frames: 1t1b 2t2b 2t4b 4t4b 5t5b

So you can restore frames 1 2 4 and 5. But frame 3 doesn't exist in the video.

OK, Thank you for telling me this. But can I use MVTools to get perfect motions on that missing frame, for example if I tweak the settings in the plugin for more accurate motion detection

[jagabo]

can I use MVTools to get perfect motions on that missing frame, for example if I tweak the settings in the plugin for more accurate motion detection

You might be able to tweak the settings to get better results but you will never get perfect results. MvTools (and all the motion interpolation tools I know of) uses looks for motions of rectangular blocks of pixels. It can't handle complex motions and morphs. Take for example an open hand that closes into a fist from one frame to the next, viewed from the side. You and I know that an in between frame should show the fingers curled on their way to becoming a fist. But MvTools doesn't know it's a hand and that fingers curl. It will probably show a 50:50 blend of the two frames, something that looks like a double exposure. See the video in this post for a bad case example:

https://forum.videohelp.com/threads/339017-Motion-Interpolation-%28VidFIRE%29-Software?...=1#post2107214