Have you tried the 24PA pull-down setting in Interpret Footage? Not the regular pull-down removal option, I mean.
I tried every single pulldown-method. Normal 3:2 and 24Pa each with five field orders (the german version says "24P Modus" in the pulldown menu for the lower five opportunities but I think it's an translation issue and stands for 24Pa). I read on the internet that WWSSW should be the right pulldown method for the gh1 and that's what AFX guesses. But it doen't work.
If I click on Guess 24Pa Pulldown it does nothing at all ....
Have you tried figuring out whether or not the clip has pulldown applied? Do this by interpreting the footage at full frame rate (29.97 for ntsc) separate fields, then either drag it into a comp with double the frame rate or Alt/Option + double click on the footage in the project window to open the footage in a footage window and step through it a frame at a time. If it has a pulldown scheme applied you'll see some duplicate frames in some kind of uneven cadence. If the footage is just interlaced you won't see duplicate frames, but if the interpretation is wrong the motion will go back and forward on each frame. If the footage is progressive then you'll see pairs of duplicate frames.
The other thing you should know is that each shot must be interpreted separately. If you have a clip with more than one shot on it there's only a small chance that both shots will start on the same frame. IOW, you may get aabba in one shot and aaabb in the other.
I hope this helps.
The footage has definitely a pulldown applied because the output of the gh1 is specified like that and other programs like neoscene and voltaichd do the pulldown perfectly but with recompression like I wrote earlier. And the clips I used to test this always consists of one shot because they come directly from the camera.
And there is absolutly no structure behind the broken frames. One time it's e.g. frame 5, the other time #6 and 5 is ok again - with the same footage and the same settings. It seems that AFX is overburdened by the footage and sometimes, it gets it right and other times not.
I suspect this issue is not strictly related to pull-down removal, but it's perhaps related to some decoding issues (shuffled/out of order frames) with recent AVCHD flavors which appeared after CS4.
Try converting a short sample to another format (Quicktime, AVI) and then import that in AE and see if you can get properly remove pull-down.
That sounds pretty convincing.
I tested a few things including your advice with the following results:
I exported the file directly without removing pulldown to a quicktime file in animation format. I'm not completly sure how to preserve the field distribution exactly but I chose upper field first on the input as well as the output settings and that should do the job. When I reimported this quicktime file back to after effects and compared it to the original avchd file in the footage-window it became obvious that the frame order is different. When I go through the original footage slowly frame by frame it is distributed nicely in 2:3 which is not the case for the just created quicktime movie. Actually, there is no structure at all in the quicktime sequence, I wrote it down for the first frames and came up with 2:5:2:1:2:5:2:3:2:3:2 ... Of course, the pulldown doesn't work with this file either as a result of that.
Something that is really interesting is that when I analyse the original footage slowly frame by frame, it has the right frame distribution like I mentioned. But when I scrub through the footage by mouse fast so that parts of it are written to the ram and then analyse this section again, the sequence is wrong in a manner like the quicktime file. After that, when I empty the cache manually and go throught the same section slowly, it's correct.
I don't have an explanation for this but I think that if AFX has to output the frames quickly for rendering or fast scrubbing it ends up with wrong frames, but if it has time, everything is correct. It would be interesting if this is a problem with my system or After Effects + avchd in general.
edit: I tried the conversion avchd -> quicktime with Adobe Media Encoder and ended up with a similar random field sequence: 2:3:1:1:1:4 ...
Somehow I missed that the footage was AVCHD. That explains a lot. I wrote an article a long time ago about the problems of working with AVCHD footage. You'll find it here.
Since that time I've had several occasions to work with the footage. None of them have been by choice. The problems pop up more often than not. They don't seem to be related directly to the different cameras. They seem to be more related to the motion in the image. I'm sure that they are caused by the MPEG GOP compression scheme that squeezes HD frame sizes into a data stream smaller than DV. That's 6 times the number of pixels at a full megabit per second less data than DV (24Mbit/sec vs 25Mbit/sec). If the hardware that's compressing this video on the fly has any problems at all you'll end up with fouled up data. There's no way around it. There's no easy fix.
Wikipedia lists several solutions for converting AVCHD to a usable format for editing. You might look at www.shedworx.com/. They have more than one possible solution. The bad news is that I've never run across a solution that works every time.
Sorry, I forgot to mention that the footage of the GH1 comes as avchd.
It seams that avchd is quite a difficult compression format but I think as other application can handle it, AFX should be able to do the same. At least for final renders. I understand that it's not possible to have quick processing and previewing with avchd files but I can deal with that using proxys. But as long as the render output is useless, there is no way at all to have a pipe through after effects without recompression.
As I said, I tried voltaichd from shredworx that you recommended but the loss in quality is not acceptable.
... as long as the render output is useless, there is no way at all to have a pipe through after effects without recompression.
As I said, I tried voltaichd from shredworx that you recommended but the loss in quality is not acceptable.
If you render to a lossless codec you're not really re-compressing, you're re-encoding. I never re-compress a lossy format as part of my workflow. If I have DV or HDV footage as the original source and the footage is not going directly back out to tape in the same format I always render to a lossless or nearly lossless format. This is especially true if I'm going to DVD or Blue Ray. You'll get much better re-compressing for either of these disk formats if you give your authoring or compressing software an lossless format to start with. IOW compressed to compressed to DVD = bad. Compressed to lossless to DVD = much better.
The only time I render to a compressed format is when that footage is going right back to the same tape format with no further processing. In most cases I don't even do that. If I'm going to do some serious work on the original video DV original footage is always transcoded to animation or at least 8 bit uncompressed before keying, scaling, or major color correction. HDV is always reprocessed to at least DVCPro 100 and the frame size is changed to HD. You can't fix compression errors by saving your video in a lossless or nearly lossless codec, but you will get much better results in the long run.
If I have any motion tracking or roto work I always separate fields on the original, double the frame rate of the appropriate square pixel comp for the format, and then export to a lossless format. I don't end up with video that is intended for playback, I end up with video that's intended for re-work. If the footage is heavily compressed like HDV or DV I use something like Magic Bullet's Deartifactor or the RE:Vision Effects Smooth Kit and Fields Kit to clean up the artifacts and provide the highest quality deinterlacing.
Every time I've successfully used AVCHD footage for complex work I've followed this workflow. If you've found a solution that will successfully separate fields on your original footage then I heartily suggest that you export the footage from that app in a lossless or nearly lossless format before you do any other work on the footage. Personally I've never found AVCHD footage with fields or frames fouled up that I could repair. I think the fouled up fields and frames happen when the footage is encoded by the camera.
If you've found a solution that will successfully separate fields on your original footage then I heartily suggest that you export the footage from that app in a lossless or nearly lossless format before you do any other work on the footage.
That's the point. I found none. Processing lossless is self-evident but thats exactly my question: to get it into After Effects without recompression.
And as far as I can tell there arent frames fouled up in the footage. After Effects fouls the frames, that's what I'm trying to explain.
edit: If somebody wants to give it a try: here are a few footage files in 24p (->60i) and 60p native. The 60p files work without problems. Unfortunately it's about 550MB to download: http://www.dvxuser.com/V6/showthread.php?t=170784
I looked at the footage and the problem is exactly as I described. Take the 00009.mts footage, don't separate fields, and you'll see some bad frames. Separate fields and you'll find odd cadences, and guess pull down and the frame rate will be correct but there will be frames out of order.
I see it in any app that has the ability to separate fields and view the footage a frame at a time.
The footage tends to look ok when played at speed, but if I look at it a frame at a time the problems are all there.
I looked at the 00009.mts element and to me it seems that every frame is there and in order. Could you give an example of missing or frames out of order (perhaps by frame-number)? And if you don't separate fields on a pulldowned clip there are of course bad frames. There are always 3 good frames and 2 bad frames where the fields are visible, that's exactly what I would expect on every video with pulldown applied.
Which apps did you use to look at it frame by frame? I have only After Effects and Premiere available. AFX is often not able to provide the correct frames but in the source monitor of premiere, everything works fine. But thats enough proof to say that the program is the problem and not the file.
Finally, I found out what the problem is! After Effects isn't able to separate the fields of an (GH1-)avchd clip. If you separate the fields either by removing pulldown or not, the playback and the export is broken. If you import it as progressive video, than export it e.g. as a tif sequence (with separate fields again) and than reimport it and remove the pulldown everything works perfect. Exept the case that you lower the comp resolution. If you do that, you get fouled frames again but that happens even with clips completely created by after effects where you have a pulldown applied at rendering.
So there is a workaround without recompression in after effects but it contains two render-procedures if you want to work in lower comp resolutions (1. encode to another format with progressive interpretation and 2. remove the pulldown) before you can start to work. Thats quite a waste of time and disk space.
AE, Final Cut, Premiere Pro...
Separate fields only, double the frame rate of a project and step through and you'll see a fouled up frame order. Start at 344 and you'll see this:
AAABB AAABB then at 154 AAAB then at 158 AAAABB AAABB AAABB AAABB
IOW, the cadence is all fouled up starting at 154 with only 1 B frame, then there are 4 A frames, then it goes back to normal 3-2 for a while.
This has nothing to do with the way the AE or FC, or anything else removes pulldown. It has to do with the fouled up 3:2 pulldown the camera added to the footage. This is what gives you funky frames when you separate fields and remove pulldown.
As far as loosing quality, if you don't re-interpret the footage by changing size or fields you get exactly the same pixels that the coded decodes and puts on the screen in AE. After Effects works uncompressed with whole pixels natively. If you render to a lossless or uncompressed you get exact duplicates of the original pixels (baring any gamma shift due to codec differences). There cannot be any loss in resolution or additional compression artifacts added to the footage.
So, what's the solution for AVCHD? Don't shoot 24P... The AVCHD pulldown scheme is fouled up. The only way to fix it is to go into the clip, separate fields, and start duplicating and removing frames to get them back into the correct cadence.
You tell me a lot of things I already know for a lot of years, but as far as this issue is concerned, you are wrong.
Have you tried the workaround I described? Doesn't it work?
I looked at the footage and starting at Frame 154 the cadence is ABBBA ABBBA ABBB and so on. Everything is right in the footage, the problem are the apps that you are using.
To proof it, I converted a few frames (150-166) to a jpg-Sequence using the method described above. You can either look at it with separate fields and you will find AABBA AAB.BA AABBA AABBA AABBA AABBA AAB (the point indicates where 154 starts - the cadence is the same as above) or you can remove pulldown using WWSSW which works also. It's exactly the part of the avchd sequence, encoded to jpgs by after effects. You only have to import the avchd as progressive.
The compression artifacts are ugly but I had to use jpg because max. attachment size is 5MB.
If you look at the jpg sequence using a picture viewer you will find 3 good frames and 2 bad frames in turns because the field order written above is combined into pictures like this: (AA) (BB) (AA) (AB) (BA) (AA) (BB) (AA) (AB) and so on. AA and BB are complete frames as in the 24p-stream, BA and AB are combined frames and look bad. Everything as expected.
150-166.zip 4.7 MB
I think you're missing the point. Separating fields in any app that will decode your sample footage gives you odd cadences. Separating fields doesn't add frames or change order, it just combines fields into complete frames. You can view the individual fields by doubling the comp or sequence frame rate.
When you try and remove 3:2 pulldown from a clip with an uneven cadence you get fouled up frames. There's no way around that. I've been removing 3:2 pulldown since AE 3.0 for everything from national TV spots to Imax movies from almost every codec that you can pull into a NLE or compositing program. AVCHD from several cameras I've tested simply doesn't properly add the pulldown when it's encoding the footage in the first place. Occasionally I run into some AVCHD footage that does it right. I know of no way to fix the problem when the cadence of the interlaced frames does not follow an even pattern.
Here's another way of looking at the problem. If 24P footage from your particular source is fouled up in After Effects, Premiere, Final Cut, Shake, (on the Mac) and After Effects, Premiere, Vegas on the PC, and the field order and cadence is wrong in all of these apps when you separate fields, then there's probably something wrong with the way the 3:2 pulldown was added in the first place. The only systems I have experience on that I haven't used to look at your problematic clip with is Avid and Nuke. The only way to change my mind is to show me an app that will show me a uniform pattern in the fields from this sample footage.
A bit confuse with your explanation. Probably because I am not too expert. This became more stress for me since I am work with Mac. I just need to get used to it. Thanks anyway
The only way to change my mind is to show me an app that will show me a
uniform pattern in the fields from this sample footage.
Do you think I manipulated the jpg-sequence to insist on my point of view? Try it yourself: import the 0009-footage in AFX, interpret it as progressive and render it out to an image sequence. You will get exactly the frames I uploaded (and a few more). Please, try it yourself. If your output is different from mine, I would be interested to hear about. And if it doesn't: what's you explanation for this?
The point is: on the one hand everytime you don't seperate fields and render you get the right sequence. On the other hand if you separat fields and render, the output is completly random. Render a few frames, empty the cache and render again. You will get different frames. With that in mind, how can you say that the apps work properly?
And please, stop talking to me like to a little child. I know how to look at separate fields and that if the cadence is wrong, the pulldown ends up with wrong frames. What I have been saying for a few posts now is that every frame is in the footage and in accurate cadence.
No I don't think you manipulated the footage. You just didn't understand my explanation. You also made my point. Interpret as progressive (or separate fields and render progressive) with field cadence fouled up and you'll get funky frames. There's no way around it. Every place there's an AB pair the interlacing artifacts will show up. If I were to interpret progressive I'd get the same results you have.
I'm not trying to talk to you as a child. Just pointing out that there's something wrong with this footage in AE, FC, Vegas, Shake, and Premiere. Therefore it's my opinion that the footage was not right when it came out of the camera. Every app shows the field order fouled up. The only way to get frames out of this footage that don't have the interlacing artifacts is to separate fields, double the comp or sequence frame rate, and then render progressive. This will give you twice the number of frames, but there won't be any AB field artifacts.
Interpret as progressive (or separate fields and render progressive) with field cadence fouled up and you'll get funky frames.
No!! You are wrong here. You will always get funky frames when you interpret a pulldown-sequence as progressive. I respect your knowledge but at that point you are blunted by habit (sorry, I don't know the correct expression in english). Create an animation in afx (e.g. a bouncing box), render it with a pulldown applied in the render settings and than import it as progressive. What do you see? You'll get exactly the 3 good - 2 bad rhythm we have in the jpg sequence and I explained why in one of my last posts.
You also didn't get my point that it makes all the difference wether or not you interpret it as progressive. If you do so, the export works fine every time you render. If not, the output is random. And I mean random! Render the same section with the same settings twice and you will get different results. You will have to empty the cache or restart afx in between, though. You can't explain this behavior by a fouled cadence. If it was just that, you would get funky frames, but everytime the same funky frames.
Gnurpf: Please try to remember that Rick is a person who is giving from his own time to help you (and others).
Sometimes these things are painful to diagnose or even put into words though the Internet, and it's very valuable that somebody is willing to sit and go through it all, even if it ends up being something else.
So what is your point? Should I just agree with him?
My last post might look a little rude at first glance but we are not going further for quite a time now and I try to emphasize the important aspects of my writings so they don't get ignored. So I use all this bold, italic, exclamation-point ... stuff to make my point clear.
A simple solution to this the problem of removing pulldown from the sample 24p clips in this thread doesn't exist. It doesn't exist because the pulldown is not uniform throughout the clip. It's as if the individual shots were edited. In order to successfully remove pulldown the order of the fields must either be a uniform pattern of three and two fields (24p WWWSS) or four and one field (24Pa WWSWW) throughout the clip. The pattern can begin on any W or S field. When clips with pulldown not removed are edited the cut can happen on any even field. Each new cut may begin with either a W or an S field.
The sample provided clearly shows a pattern of WWWSS and then occasionally a WSWWW pattern. Most of the time the pattern is 24p (three W's and two S's) but every once in a while there's a single 24Pa (four W's and one S) pattern. It looks like the footage has been cut or there is an occasional missing field.
Inadequate processing speed inside the camera that results in dropped fields could be one explanation. It's the explanation that makes the most sense to me. Another possibility may be the GOP compression of AVCHD. I don't like this explanation as much because I've seen some AVCHD 24p footage that works. The codec decoding software common to all of my tools on Mac and PC may also have a problem, but this is the least likely suspect.
To replicate the problem I took some known 24p footage from a NatGeo special that I shot on a few years ago on film. The project was cut on tape in an online suite (before NLE's). I imported a bit of this tape and then checked the fields. There were places in the footage where a 3;2 pattern was followed by a single 4;1 pattern that then went back to a 3;2 pattern on a scene change. The footage behaves exactly the same as the sample footage. Any attempt to produce non interlaced output results in random ugly field artifacts.
The detailed descriptions above were intended for anyone following the thread that may be confused by this very confusing issue.
It doesn't exist because the pulldown is not uniform throughout the clip.
The pulldown is uniform throughout the clip!
I appreciate that you spend your time trying to help me! But as long as you write text blocks instead of responding to what I wrote you doesn't help at all.
In the sample 00009 that you provided if you start at field 344 you get the following frame grouping:
WWWSS WWWSS WWWS WWWWSS WWWSS WWWSS
In the third frame the last S field is dropped. In the fourth frame there is an extra W field, then the frames return to the normal pattern of WWWSS fields.
I checked this again and again in several apps. It's always the same. I wouldn't make this up. I've seen this before. I'll probably see it again.
Sorry that I don't have a solution for you.
1. Import the footage - progressive frames
2. Export to an image sequence
3. Import again - separate fields now
4. Have a look at frames 334 and the following
5. Feel proven wrong.
This is the last thing I'm going to say on the subject.
If you interpret interlaced footage as progressive you will not get or render fields when you render.
'5. Feel proven wrong.' How lame! This isn't Friday Night Smackdown you idiot. This board isn't about being right or wrong, but to share information and help with people who work with and love AE. Rick was willing to take the time (10 times) to share his knowledge, experience, and insight on the subject with you and everyone else on the board. Thank you Rick. But you came here looking for a fight from the beginning. If you are that damn smart, why are you here?????? If you want to fight about 3:2 pulldown, get a hold of Eugenia at eugenia.gnomefiles.org. She knows as much on that subject as anyone on the internet, and she loves to fight. Tell her Cindy sent you. Oh, and by the way, it will be like her having a battle of wits with an unarmed person. Good luck.
Oh, yeah ... I like this kind of forums.