Choppy Movement After Encoding With Xvid

[Amaze]

my newly compressed video is kinda choppy when there's a lot camera movement. the original source(picvideo) is not like this. I used virtualdubmod's resize-filter, logo-filter, and the smart deinterlace-filter. I recorded the source in 29.967 fps, but I only found one fps-setting in the xvid-settings and it looked like it was for calculation purposes only. so I don't know what really caused this. any ideas?

[Shadowmistress]

Just curious, but what fps does virtualdub report the final clip to be?

[Amaze]

not sure, where is it shown?

[redwudz]

In VD, File>File information. You should see the FPS and the Data Rate of the file.

[jimmalenko]

File > File Information

[Amaze]

it's 29.970

[junkmalle]

If your source was film inverse telecine instead of smart deinterlace. Smart deinterlace will leave you with choppy results with telecined movies.

[Amaze]

it was just regular tv, what exactly is inverse telecine?

[junkmalle]

In short, inverse telecine will restore the original 23.97 fps progressive film frames from a movie that has been broadcast as 29.97 fps NTSC television.

If your source was originally on film you may be able to do this. If your source was a live TV broadcast (like a football game) you can't.

Here's the easiest way to tell if your source file is telecined film: find an scene where there's stuff moving and step through frames one by one with VirtualDub. If you see a repeating pattern of 3 clear frames followed by 2 frames with interlace artifacts you have a telecined movie. (Other patterns are possible but it's almost always this 3,2 pattern.)

How jerky is your Xvid file? A Smart Deinterlaced telecined movie will have 6 jerks per second because one frame out of every five is a duplicate.

[Amaze]

it's only jerky in the beginning, coz the camera sweeps over the audience. but I couldn't tell anything later on when the camera was still...I looked for the pattern, you were right, 3,2...

[Amaze]

is there any cure?

[junkmalle]

is there any cure?

Inverse Telecine (IVTC).

But to get perfect results you need to have no dropped frames and your source has to be perfectly telecined (no changes in the 3:2 telecine pattern). Otherwise it will require a lot of work to get perfect results.

VirtualDub has its IVTC settings under Video -> Frame Rate. At the bottom of the dialog you'll see the inverse telecine section. You can try the automatic setting "Reconstruct from fields, adaptive" but I find it doesn't give perfect results. Which leaves you with the manual settings. I always use "Reconstruct from fields, manual".

Along with that manaul setting you have to set the Offset value and Polarity. I've never had to checkmark the Polarity box so you probably won't either.

There are 5 possible settings for Offset, 0 to 4. You can just try each value until you find the one that works: you can tell when you have the right value by watching the output image while saving your AVI or with File -> Preview Filtered. If there are no interlace artifacts you have the right value. If you have the wrong value you will see interlace artifacts now and then.

Note that the program doesn't IVTC while you move the frame slider around or step through the source with the arrow keys. You have to actually save or preview to see results.

The other way to determine the correct value is to look at the input image as you step through frames. You count the number of progressive (non-interlaced) frames from the starting frame to the first interlaced frame. The number of progressive frames is the value to enter as the offset. For example, using P to represent a progressive frame and I to represent an interlaced frame, you'll see one of the following patterns:

IIPPPIIPPPIIP... 0
PIIPPPIIPPPII... 1
PPIIPPPIIPPPI... 2
PPPIIPPPIIPPP... 3
IPPPIIPPPIIPP... 4

The number on the right is the offset value to enter for that pattern.

If you have a still (very little motion) section at the starting frame you won't be able to see the interlace lines. But remember the pattern repeats every 5 frames. So you can zip ahead 5, 10 100, or 3245 frames and start counting there.

The problem with the manual method is that you often have to encode in sections. Any time there is a break in the 3:2 pattern you have to start a new section.

What I usually do is determine the value to use at the begining and save the output until I see interlaced frames in the output window. I then stop the encode, start a second instance of VirtualDub to examine the output file and figure out exactly where the interlace lines started showing up. I then save the clean section (from the second instance) with Direct Stream Copy, and go back to the first instance and mark-in at the exact frame where the telecine pattern broke (ie, the first frame after the clean section) and start encoding again. Of course you usually have to enter a new Offset value for the new section. I repeat this procedure until I've got the whole film IVTC'd. Then I put all the pieces together with VirtualDub's Append AVI Segment feature and save with Direct Stream Copy.

In my experience, about 1 in 10 recordings can be IVTC'd in a single section. 8 in 10 require 2 to ~20 sections. If there are too many breaks I just give up and use another method (none of which gives perfect results).

Sometimes you'll find that the telecine pattern breaks with almost every new camera shot. This happens when a movie is shot on film, telecined, then edited as video. Some modern made-for-tv movies are made this way.

[Amaze]

looks like a lot of work, thanks for your help.

[Amaze]

I'm gonna try the "Reconstruct from fields, adaptive" first, to see the result. can I do it while encoding to xvid or do I need to do it by itself?

[junkmalle]

I'm gonna try the "Reconstruct from fields, adaptive" first, to see the result. can I do it while encoding to xvid or do I need to do it by itself?

You can do both at the same time.

[Amaze]

this is the first time I've encountered this reverse telecine thing, and it's also the first time I've recorded from digital tv. I hope all digital programming isn't like this...

[junkmalle]

I hope all digital programming isn't like this...

It is. Anything you capture via a composite, or S-video cable will be interlaced -- that's all you TV can display. The exception might be cards that can capture the digital MPEG stream directly from the cable company or satellite. Movies captured that way might be progressive 23.97 fps -- but I don't really know since I don't have one.

If you were going to burn to a regular movie DVD you wouldn't have to worry about any of this.

[Amaze]

yea, I know it'll all be interlaced, I mean this telecine-thingy. is all digital tv like that?

also, should I enable inverse telecine during the first pass? I always enable everything during the first pass like filters and all that, just wondering if it's unneccesary and slows me down?

I just did the reconstruct from fields adaptice direct stream copy and it's exactly the same.

another thing I'm a little confused about is when I use the inverse telecine option, do I still need to use the smart deinterlace filter?

edit:
ok, now I tried manual inverse telecine, none of the offset values removed the artifacts in the preview window.

[junkmalle]

yea, I know it'll all be interlaced, I mean this telecine-thingy. is all digital tv like that?

Movies are filmed at 24 frames per second. To display them on a TV that runs at 59.94 fields per second they have to display frames more than once. They slow the film down to 23.97 frames per second and display a frame for 2 fields, then next for 3 fields, and repeat this over and over. So, on average each frame is displayed for 2.5 fields. If you multiply 2.5 * 23.976 you'll see the result is 59.94. This is called "3:2 pulldown" or telecine. Every movie you record will have this sort of telecine pattern. Live TV, things like football games, will be fully interlaced, not telecined.

also, should I enable inverse telecine during the first pass? I always enable everything during the first pass like filters and all that, just wondering if it's unneccesary and slows me down?

Yes, both passes have to be identical. Consider using single pass Quantization mode if you don't need your files to be an exact size. This is essentially a constant quality mode. You pick the quality you want and the file comes out to whatever size is necessary to get that quality. The results are as good as two pass but it only takes one pass. The downside is you don't know exactly how big the file is going to be. But you can watch the projected file size in the status window to get an idea how big the file is going to be.

I just did the reconstruct from fields adaptice direct stream copy and it's exactly the same.

Of course, direct stream copy turns of all filtering and processing. All you did was copy the source file!

another thing I'm a little confused about is when I use the inverse telecine option, do I still need to use the smart deinterlace filter?

No, you use IVTC instead of deinterlacing.

edit:
ok, now I tried manual inverse telecine, none of the offset values removed the artifacts in the preview window.

Are you still using direct stream copy?

[Amaze]

thanks for taking the time. as for the preview filtered, it says "dub in progress" in the window, should I wait for it to finish? coz I just stopped it after a while and went back a few seconds to look at the frames.

[junkmalle]

thanks for taking the time. as for the preview filtered, it says "dub in progress" in the window, should I wait for it to finish? coz I just stopped it after a while and went back a few seconds to look at the frames.

Until you get your technique perfected use a single pass compression. With single pass you can use "Save As AVI" to start a conversion, then stop it any time to examine the resulting file.

[Amaze]

worked like a charm, thanks for your help. it was very informing.