Track an extended bicubic surface

G’day mate!

I have an animation that should be on glasses, but the glasses are made of a continuous surface like Doc’s glasses in the film Back to the Future II. I tracked the animation on the glasses and the track also looks very good now comes my question … I then bent the animation about extended bicubic so that it adapts well to the glasses … unfortunately it is not possible to track the surface by extended bicubic of the glasses there are an unbelievable amount of reflections now I would have to bend the animation frame by frame so that it adapts to the glasses is there a solution to circumvent this?

This Flame Learning tutorial might help.

1 Like

thanks for the tip I’ll have a look at this right away

If you’ve tracked the glasses already (which is how it reads to me) can you use that to stabilise the plate then bicubic your stuff on top of that then apply the track?

@johnt unfortunately not, I tracked different areas of the face with masks that are attached to an axis. As far as I know, it is not possible to stabilize this … but if I am wrong I am open to ideas.

Track it with any thing. Even a one point track. Then you can hand animate on top and it will become easier. Split up the problem into smaller pieces. If it rotates around then split the shot up into sections like frame 1-15, 16-25. Whatever it takes. As a wiser man than me once told me: there’s no magic button Johnny.


Just remember to check the inverse stabilise works before doing any work.

1 Like

at the beginning i tried to stabilize the shoot but then after a short time i got to tracking but you are right i give the stabilization a second first try, what might also work would be a geotrack

It probably would but I’m pretty sure you can track in flame quicker. And you’ll feel like a hero when you e done it.

I fully agree with you i would like to solve this in flame and the world needs heroes.

FWIW: You can track a bilinear, then change it to a bicubic. The 4 corner points will retain the track and the tangents will follow along properly. You can even subdivide and move the points around and for the most part they will retain the original 4 point tracking info.

1 Like

@ytf I’ve already tried that, the problem is that I have to bend the surface manually over 300 frames. The movement is unfortunately very strong.
I am not 100% satisfied with the result.
I will now try to stabilize the shot as well as possible to give it a try. Maybe it will work.
Thanks for the help, I appreciate that!

I always go the stabilize route myself. Sometimes I do what I can with 1 or 4 point, then I use the MotionVectors on top of that for the loosey goosey stuff, then re-apply the inverse of the 1 or 4 point. Of course, not actually seeing your shot, I have no idea if that would be helpful.

So the glasses are like this?

Sounds like possibly one way of thinking about the issue you’re having is that the bicubic is bent, yes, but it’s really only distorted in 2d and/or the corners are not sitting in the correct depth. You could try a camera track and then snap your points of a bicubic to some strategically placed locators (or just eyeball it based on the point cloud) on the glasses and then use a diffuse map on that surface. I wouldn’t use the flame camera tracker for this though, so something like Syntheyes or PFTrack or whatever flavor of matchmove software. If that’s not available though, I’m +1 for the stabilize approach Tim and John are suggesting.

Also use a grid as your front and add blend mode surface in action. Crok grid or something is good. As is sapphire grids. Try not to have many subdivisions in the bicubic otherwise it gets a bit confusing. And don’t forget you can copy and paste the key frames in the animation channels (using the copy and paste buttons on the right hand size very useful if you fuck it up and need to get back to unity).

1 Like

Also always remember this:

If it looks right, it is right.


I’d build and track geo and use reflection maps.

1 Like