16fps w/ pulldown looks awful on finished dvd

[henrybirdseye]

I'm working on a project involving a lot of 16mm film shot and transferred at 16fps... so there are a lot of jitter frames. I exported .m2v files using Cleaner. If I watch the encoded m2v's, they look great, the fields are separated the way they should be.

BUT, the problems start when making the dvd's. I created a test disc using DVD-Lab Pro, which requires assets video of m2v's. When I watch the finished disc, the fields have somehow gotten scrambled or interpolated in some way and the video look bad, very bad.

Same result when authorinig using Encore 2.0. How can I tell these programs not to futz with the fields?

HB

[SLK001]

How can I tell these programs not to futz with the fields?

You can send each of them a strongly worded email.


Why don't you change the framerate to 24fps, then apply a normal 2:3 pulldown? It will be sped up by 33% and may not be all that objectionable to view.

[henrybirdseye]

The client wants to preserve the framerate at 16fps. Besides, then I'd have to remove the odd pulldown sequence or it'd go REALLY bad,

[jagabo]

Perform a hard pulldown and encode as 29.97 fps interlaced?

In theory you could use a simple frame duplication to convert 16 to 24 fps (this wouldn't require much more bitrate than 16 fps because exact duplicate frames compress down to nearly nothing) and use a custom 4:4:4:3 pulldown (that knows some frames are duplicates and avoids them) for 59.94 fields per second. This might be enough to convince your mastering software not to futz with the video but you'd probably have to write your own pulldown program.

[SLK001]

You might try using DGPulldown's custom settings and using 16 as your input FPS and 24fps as your output FPS, then HARDCODING this using your encoder. Then apply a normal 2:3 pulldown prior to importing into your authoring program.

[SLK001]

Also, you may want to use a deinterlacer (field blender) to eliminate the interlacing lines that will naturally occur.

AVISynth would be perfect for this.

[edDV]

For a playable NTSC DVD there two rates allowed and only two.

23.976 frames per second 480p

59.94 fields per second (480i 29.97 fps)

To get to 23.976 by frame repeat, you start by running the projector at 15.9840 fps, then convert by repeating a frame every 4th frame.

To get to 59.94 fields per second, the projector still needs to run at 15.9840 fps (31.97 fields/s) but a custom telecine field repeat rate needs to be created or frame interpolation needs to be employed. Frame conversion is now 15.9840 to 29.97 or 1.8750 frame repeat. It is left as an exercise to calculate a field repeat sequence that preserves 1.8750 ratio.

[henrybirdseye]

I've come to believe that it's something in the Cleaner profile that sets flags in the resulting m2v's that cuases the authoring programs mess with the interlacing.

So, anybody know Cleaner really well?

The correct answer is indeed to pretend that it's all interlaced video, but getitng the encoder to see it that way is the trick.

hb

[jagabo]

To get to 23.976 by frame repeat, you start by running the projector at 15.9840 fps, then convert by repeating a frame every 4th frame.

That only gets you up to 19.98. You have to repeat every other frame to get 23.976.

To get to 59.94 fields per second, the projector still needs to run at 15.9840 fps (31.97 fields/s) but a custom telecine field repeat rate needs to be created or frame interpolation needs to be employed. Frame conversion is now 15.9840 to 29.97 or 1.8750 frame repeat. It is left as an exercise to calculate a field repeat sequence that preserves 1.8750 ratio.

I already gave it: 4:4:4:3

[SLK001]

What the heck is Cleaner? Is that someone who cleans your house once a week?

What is the actual encoded framerate of your video (not the PERCEIVED framerate, which is still 16FPS)?

I'm assuming that there is no audio with this, so there isn't going to be any syncing issues with the resultant video. If you use the 16->24 encode, then pulldown to 30FPS, you can "lie" to the authoring app by changing the framerate to 29.97.

[edDV]

To get to 23.976 by frame repeat, you start by running the projector at 15.9840 fps, then convert by repeating a frame every 4th frame.

That only gets you up to 19.98. You have to repeat every other frame to get 23.976.

To get to 59.94 fields per second, the projector still needs to run at 15.9840 fps (31.97 fields/s) but a custom telecine field repeat rate needs to be created or frame interpolation needs to be employed. Frame conversion is now 15.9840 to 29.97 or 1.8750 frame repeat. It is left as an exercise to calculate a field repeat sequence that preserves 1.8750 ratio.

I already gave it: 4:4:4:3

You are correct. I was just trying to go through the math. I hadn't run the numbers for 16 fps before.

[jagabo]

I don't know that 16 fps with 4:4:4:3 pulldown is legal for DVD.

[edDV]

I don't know that 16 fps with 4:4:4:3 pulldown is legal for DVD.

It will look like 480i to the player and should play fine at 480i. It may confuse the telecine detectors for progressive playback, but usually they need to detect a 2:3 5 frame sequence to trigger an IVTC. Otherwise they pass the field sequence to the deinterlacer. The deinterlacer may screw it up.

23.976 will always be considered progressive by the player but it will play a bit studdered with every other frame repeated.

For 480i playback, the player will apply 2:3 telecine.

For 480p playback, the 1:2:1:2 frames will have 3:2:3:2 frame repeats applied. It will be 50-50 whether you get 3x applied to the single or the doubled frame first

Case 1
23.976 from 15.9840
ABBCDDEFFGHH
3:2x to 59.94fps
AAABBBBBCCDDDDDEEEFFFFFGGHHHHH

Case 2
ABBCDDEFFGHH
2:3x to 59.94fps
AABBBBBCCCDDDDDEEFFFFFGGGHHHHH


Some original 16fps frames end up repeated 5x others 2x, 3x. This all may end up looking jerky or the eye may just average it out.

[manono]

For a playable NTSC DVD there two rates allowed and only two.

Not exactly true. For NTSC, any base framerate between 19.98 and 29.97fps can be pulled down to display at 29.97fps. Any framerate from 2/3 of the display rate on up. For PAL, this means 16.667fps up to 25fps can be pulled down. Just last night I encoded a 20fps (19.98) silent film and then applied pulldown to output 29.97. It plays quite smoothly at 3:3.

One solution for the OP would be to repeat a frame out of every 4 to get it up to 19.98fps, and then apply pulldown from there. Better than doing similar to get it up to 23.976fps. I don't know anything about the programs henrybirdseye's using. AviSynth can do it pretty easily.

I have seen some 16fps silent films on DVD, and there are duplicate frames every so often. I don't think there's any way to do it without adding dupe frames and making it play slightly jerky. Therefore, my add-dupes-to-19.98-and-telecine-from-there is as good a method as any, and the way I'd do it. Saves on bits also as you're applying a soft telecine, and not hard encoding it at 29.97fps. No interlacing either, encoded into the video. All progressive.

[edDV]

For a playable NTSC DVD there two rates allowed and only two.

Not exactly true. For NTSC, any base framerate between 19.98 and 29.97fps can be pulled down to display at 29.97fps. Any framerate from 2/3 of the display rate on up. For PAL, this means 16.667fps up to 25fps can be pulled down. Just last night I encoded a 20fps (19.98) silent film and then applied pulldown to output 29.97. It plays quite smoothly at 3:3.

One solution for the OP would be to repeat a frame out of every 4 to get it up to 19.98fps, and then apply pulldown from there. Better than doing similar to get it up to 24fps. However, for optimum smoothness, better might be to hard telecine it to 29.97fps, as edDV is describing. I don't know anything about the programs henrybirdseye's using. AviSynth can do it pretty easily.

I have seen some 16fps silent films on DVD, and there are duplicate frames every so often. I don't think there's any way to do it without adding dupe frames, and thus, making it play slightly jerky. Therefore, my add-dupes-to-19.98-and-telecine-from-there is as good a method as any, and the way I'd do it. Saves on bits also.

I should have said 23.976p or 29.97i after all the dup field/frame efforts.

To optimize 16fps film, a science fair project is required to find the smoothest visual playback on a standard DVD player. This should be done separately for 480i and 480p, then those results compared and weighted for players and player cost.

Oh and toss in 18fps as well since that was also popular for 8mm and 16mm film (mostly consumer 8mm).

There are two issues here to consider.

One, what gets the best perceived playback. Second is if there is archive value with a clear reconstruction path to the original 16fps (or 18fps) frames so that future better technology can be applied.

[edDV]

Of course, the quick solution is to convert one framerate to another in the interpolation mix master. This may or may not work but the process does destroy any reconstruction path to original frames.

[henrybirdseye]

I wish I'd been able to have all this footahge xfered at 30fps, then I could repeat on twos and run it all at 15fps then keep shut up about it.

[manono]

I should have said 23.976p or 29.97i after all the dup field/frame efforts.

I understood what you were saying. There's only one display rate, and that's 29.97i. There are an infinite number of base framerates, however, of which 23.976p is only one. Like I said, I just made a backup of the 19.98fps new Criterion DVD of Pandora's Box just last night:

http://www.imdb.com/title/tt0018737/

Just because the DVD production companies insist on hard telecining (encoding for 29.97fps interlaced) every base framerate other than 23.976p, doesn't mean it's not perfectly possible and perfectly legal or compliant to soft telecine using patterns other than 2:3. Like I said, I encoded that movie at 19.98p, and then applied DGPulldown for 19.98->29.97fps. Plays smooth as silk at 3:3, and more smoothly than something to which 2:3 (or 3:2) pulldown has been applied.

Of course, the quick solution is to convert one framerate to another in the interpolation mix master.

Hehe, quick, maybe. Lousy, for sure, as it'll blend it all to hell.

I wish I'd been able to have all this footahge xfered at 13fps, then I could repeat on twos and run it all at 15fps then keep shut up about it.

It's not too late. It's easy to slow it to whatever you want. 14.985fps would be better. It's closer to the original framerate, you can repeat each frame once, and it'll then be at the NTSC display rate. If there's already audio in it, it'll have to be slowed as well. If not, it's a piece of cake:

AssumeFPS(14.985)
ChangeFPS(30000,1001)

[jagabo]

Or slow the frame rate down to 15 and duplicate every frame. That's only a little worse (and the other direction) than the usual PAL speedup.

<edit> Crap, manono beat me! </edit>

[jagabo]

For NTSC, any base framerate between 19.98 and 29.97fps can be pulled down to display at 29.97fps. Any framerate from 2/3 of the display rate on up. For PAL, this means 16.667fps up to 25fps can be pulled down.

Can you explain where the 2/3 limit comes from? Is it because the max number of fields for which a frame is specified to be pulled down is limited to 3? Or is the limiting factor something else?

Obviously, when you pause a DVD player it is performing an infinite pulldown!

[manono]

Is it because the max number of fields for which a frame is specified to be pulled down is limited to 3...

Yes, exactly. You can repeat each first field (RFF flag) only once. The max is 3:3 pulldown, which means the framerate can only increase up to 50% (19.98->29.97), or put the other way, the lowest framerate which can be pulled down is 2/3 of the display rate.

If you look at a D2V of standard 23.976->29.97fps 3:2 pulldown, it goes 0 1 2 3 0 1 2 3, the typical sign of 3:2 pulldown having been applied.

If you look at the D2V of something which has gone from 19.98->29.97, it goes 1 3 1 3 1 3.

[jagabo]

Is it because the max number of fields for which a frame is specified to be pulled down is limited to 3...

Yes, exactly. You can repeat each first field (RFF flag) only once. The max is 3:3 pulldown...

Thank you manono.

[manono]

I probably shouldn't stopped where I did, as I left a couple of things unsaid.

If you look at the D2V of something which has gone from 19.98->29.97, it goes 1 3 1 3 1 3.

And those 1s and 3s mean that every frame has an RFF (Repeat First Field) flag, so every frame (or 2 fields) outputs 3 fields on playback, increasing the framerate by 50%. There's no such thing as a flag that causes 2 extra fields to be output from a frame, so 50% is the max. Although you probably know these things already, quite a lot can be learned from the DGIndex FAQ, including:

The remaining numbers of the file is a reference to how the frames should be interpretated with reference to top/bottom-first (tff) display and repeated fields (rff).

trf = 2*tff + rff
0 0 0 (bottom first without repeated field)
1 0 1 (bottom first with repeated field)
2 1 0 (top first without repeated field)
3 1 1 (top first with repeated field)

http://forum.doom9.org/showthread.php?t=87809

The standard 3:2 pulldown has the RFF flag on only half the frames, so its output framerate only increases by 25% (23.976->29.97fps)