What is the Aperture width and height setting?

On the 3D Camera Tracker node, there are the settings "Aperture width” & “Aperture height”. What do these settings mean?
I initially thought that they were the settings for the sensor width and height, but I believe that’s the “Film Gate” setting. What is the Aperture width and height?

Screenshot 2023-09-18 at 4.01.49 PM.png (112.58 KiB) Viewed 963 times

Your first instinct was correct. If you change the Film Gate in the drop-down, you'll see those sliders update. They obviously can't put every sensor in the world in that list, though, so the sliders are available if you need to enter a filmback manually, or compensate for cropping or an undistort.

I don't know much about camera tracking. How WOULD you compensate for a crop or unfortunately using the manual adjustments? Do you mean letting it auto track and then correcting the numbers to re-track?

I don't know much about camera tracking. How WOULD you compensate for a crop or unfortunately using the manual adjustments? Do you mean letting it auto track and then correcting the numbers to re-track?

For crops you have the center point, which is optical center of lens in normalized coords. Set the aperture to crop area size and modify center if crop is offset from lens center.

I don’t think Fu can solve the center automatically, specialized matchmoving softwares can.

I meant sensor cropping when the camera's shoots in different formats. The Alexa Mini, for instance, shoots a 3.4k image with a fully open gate on a 28.25mm wide sensor. But when shooting UHD, the sensor is cropped to 26.4 mm. See https://vfxcamdb.com/arri-alexa-mini/

If you have an image that's been cropped in post, the best way to deal with it is to request the original unmodified clip from production. Failing that, I prefer to pad out the image to its original format and mask my trackers. That will require getting the crop and transform numbers from the assistant editor. Sometimes matchmove software can estimate the optical center from the track, but it's chancy. Better to have the highest possible quality of inputs than to try to deal with degraded footage.

If you're using an undistorted plate, the image raster size has probably changed, and you have to adjust the filmback of your virtual cameras to compensate. Here's a clip about that process I pulled from a longer tutorial (which I can't share because I was demoing on uncleared client footage):

Thanks for the in depth answer. It's hilarious that all the math you learn in Jr. High comes back around in things like this. When you're young, the type of jobs that require actual math is years away, unless you're into programing. Even then I think the more physics based problems like this and using ratios for sensor crops tend to be really satisfying.

What do you use for tracking? Mocha Pro, Synth Eyes, 3DE?

I know, right? I really wish my math teachers had known what the heck matrices were actually for. I regret not paying enough attention on that topic.

I have the most experience with PFTrack. I've done a little bit of Syntheyes. I've used Mocha for roto but not much else.