Motion Tracking results in half the solved data
-
Hi,
I'm a senior motion designer working on tracking a 518-frame shot for compositing using Cinema 4D's Motion Tracking toolset, but I've encountered an issue. The camera solve starts at frame 218 instead of at frame 0. Frames 0 to 217 lack camera movement and tracking points, despite providing sufficient manual tracks and ensuring the frame range covers the entire clip.
I'm using Cinema 4D 2025.0.2 on macOS Sonoma 14.6.1 with a MacBook Pro M2 Max (96 GB RAM). The footage is a 1920x1080 EXR sequence at 25 fps in ACEScg color space. It's a sideways trucking shot of a subject skateboarding along a building. Its a complicated shot (because of reflections) but im confident that i managed to provide solvable data.
Attempts to resolve the issue include:
- Retracking with different methods, such as using 8-bit Rec.709 footage instead of linear ACES, Increasing contrast for better auto-tracking.
- Trying a fully automatic solve without manual adjustments.
- Lowering track count, raising track count.
There were no errors during tracking or solving, and I didn't adjust any settings that could shorten the solve. The Focal Length is unknown but Constant.
Troubleshooting by solving specific frame ranges:
- Frames 0–200: Correct solve until frame 76, then stops.
- Frames 0–300: Correct solve until frame 137, then stops.
- Frames 138–300: Results start at frame 218 (note that this is the same starting frame as in the full solve).
This suggests an issue occurring between frames 138 and 218. But the incomplete solve in the 0-200 test speaks against that.
This is the first time I've encountered such a problem. Is this a known issue or a potential bug? Any suggestions or workarounds? I need to solve this urgently.
Screenshots: https://www.dropbox.com/scl/fo/rh4wg1hf0i8e2ogpv9v4n/AMz9gWYRK39KmPBJBME8LKc?rlkey=g410wfyt05jxrhphi1vwdkj47&st=4rnk5k34&dl=0
Thank you for your help!
-
Hi engineer-make,
I can retrieve only very little from screenshots. The fact that you use EXR and Aces files tells me you have converted the footage into it. But that is not something that would help the track.
It looks like you masked the reflections out.The question for me is, what camera settings were applied, what camera, etc.? Was there stabilization on the camera, or was there any post-cropping? Things like that.
I would assume that the floor has the most to offer.
Anyway, without the footage and camera data, e.g., camera model, Lensgrid, Lens data, used sensor area, etc., it is hard to say a lot.
When you take the different attempts and overlay the camera, given that the set survey is precise, do you have enough data to merge it into one camera move?
Going by the pixelated parts, I assume you can't share the footage. If so, just use HD in MP4 and via Dropbox; thank you.
My suggestion, set a Mask to get only the floor in the image. Then move in 50 frame steps and use "Create Auto Track" each time; track back and forth until you get a lot of coverage. Then solve. Without footage, this is a wild guess.
Let me know what I can do.
All the best
-
Thanks for the fast answer!
Since the original footage is RED Raw, the conversion shouldnt have harmed it too much but i also did one without converting it into linear space (by going from RED wide gamut to Rec709 2.4). I will do another test with a different file format since i have noticed some strange patterns and colors appearing when i display the footage in Tracker View instead of original.
Camera Model and lens data are sadly unknown to me and i have no tracking plate either. No post cropping but i tried unwarping the lens distortion while troubleshooting but that didnt affect the track or result.
The floor is mostly obscured in the beginning so using it wont work.
I didnt mask out the reflections or the skater, instead i selected and removed those unwanted tracks after autotracking.
I didnt try merging the data into one camera move because i couldnt figure out how to do it on the fly.I have updated the dropbox with requested mp4 ("TrackingShot"). I'll have to delete it soon though.
-
Thanks for the footage, engineer-make,
You can delete it now; I believe you do not feel comfortable with it being in public.
Was this a Komodo or a RED with a Motion Mount? Do you know if it was shot in full resolution or smaller (Like when I go down to 5k on my RED, the frame is smaller, which also means less distortion on the edges?
I will take a look now.
Cheers
-
I was at the shoot but i didnt receive any infos on the cam so im not sure. I made sure to stress that they shouldnt zoom though, so the focal length should be constant. I also mentioned that i need as much resolution as i they can capture.
-
Hi engineer-make,
As mentioned, with so little data, it gets even more difficult than without.
I stick with my suggestion; after many explorations, if you have several parts, bring the scene into scale and orientation, then match the camera. Nope, that is not how it should work.
There are pretty much three parts.
1
The first part shows a camera move with the skateboarder, which results in nearly no parallax to retrieve a stable camera path. The skateboarder takes up a lot of space; anything else is mostly reflections.
2
The camera pans and starts to move, while the focus starts to be pulled.
It is not much better than a lock-off shot, parallax-wise.
3
The best part is that the reflections are lower (size-wise) when the distance increases and the skateboarder becomes increasingly smaller.
However, parallax is not huge, and the surfaces are very even.
.I have run this with several attempts also through SynthEyes. I just wanted to see if there is perhaps a shortcoming from the Cinema 4D Tracker:
The results are similar and change based on the chosen settings (Not a good sign).As a side note, a wild guess here, in the hope it helps: The lens was a Photo lens (correct?), not a Cine lens, as I get results here that show "lens breathing" (equal to small zooming) when the focus was pulled, I feel the results split here each time; I get either the first part or the last part.
The other point is that even though I placed 50 manual trackers from frames 0-95 and had masks running from 95 to 227 while using mostly auto-trackers for this last part (I deleted all auto-trackers between 0-90), even then, I got no useable results in either app. I cleaned up all reflections and the skateboarder.
I have spent many hours on it, but the little parallax and low detail structure, while a larger portion is the missing data (reflections, etc.), and perhaps the lens breathing on top of that led to this; I might miss something, but I have no further ideas to solve it as one continuous camera move. Sorry. The key question I ahve is, what is the target idea, i.e, what will be placed into the shot? If it is just a flag, perhaps tracking the pole might be an option.
I hope someone else has a better idea. (No promise, but I have not given up! It is midnight here and I call it a day.)
My best wishes for your project
-
Hi engineer-make,
It would be the first time not to get a result, so I explored more manual solutions.
I got one! Yeah!
CV4_2025_drs_24_CMds_01.c4dPlease link the mp4 you sent me for the material.
The first part had a little bi less precision, but it quickly got better. But on an 6K monitor, it loos good.
I hope investing my Saturday into this will have some value for you, each tracking feature is manual and done frame by frame. Those results are very rewarding.It is not oriented, scaled, or touched the scene after I got that far.
I left the Tracking Marker inside the file so you see where goes what.
My best wishes for your project!
-
Hey Dr. Sassi wow i am blown away by your commitment to helping me with this problem! You are an absolute legend! Thank you so much for all the hours you put into this!
I saw your first answer on Friday and was about to thank you and ask you for suggestions on how to combine both partial solves properly (after some more tries today) and that im thankful for you sharing all the infos because it made me understand what i should have kept a very close eye on at the shoot (like having enough parallax or floor in the first half) but i didnt expect you to take your Saturday to solve that beast of a track from a simple mp4!
Im amazed!
Thank you!! -
You're very welcome, engineer-make.
Each tracking that has some challenges is a learning process, which I love. Here the key was to interpolate the features when they were obscured.
I have very little to really finish it, like measurements, but this might come close:
https://stcineversityprod02.blob.core.windows.net/$web/Cineversity_Forum_Support/2024_PROJECTS_DRS/20241103_Red-shot_tracking_04.zip
I have loaded the py version here:
CV4_2025_drs_24_MOmg_21.c4dI couldn't resist to put a figure in it, as a test.
CV4_2025_drs_24_MOmg_22.c4d
Thanks again for the kind feedback.My best wishes for the project
