< Previous | Contents | Manuals Home | Boris FX | Next >
Compositing-friendly vs Matchmoving-friendly Models
Similar to the merged vs separated models, there are “matchmoving- friendly” vs “compositing-friendly” models. In Nuke parlance, these are respectively Backwards vs Forwards models, though we’d beg to differ.
In the matchmoving-friendly direction, as SynthEyes has previously always done, if the pixel is at (x,y) in the ideal undistorted image, the distortion calculation computes where the pixel will be in the distorted image. That's what you need for doing matchmoving: you take a 3-D location, figure out where it is on the ideal camera (via the 3-D-to-2-D perspective transform), then you compute where it would be in the incoming distorted image. That distorted location can then be compared to the expected, observed, location — the location of the tracker. The matchmoving process minimizes those discrepancies (always in the distorted image, never undistorted image) by varying the camera positions and orientations, the field of view, the tracker positions, and the distortion coefficients.
That same ideal-to-distorted calculation is exactly what you need to display meshes in the camera view so that they overlay the distorted image in the correct location, and it's the calculation you need to be able to UNdistort footage (for example for lens workflow #1).
However, the historical approach goes in the other direction: given a position on the distorted image, calculate where it should be in the ideal perfect image. It's the historical approach because it was developed by astronomers who printed out their big images, measured them with caliper, and then wanted to be able to calculate where the stars would be if their telescope lens was ideal. This calculation is good for REdistorting rendered images to match the original distorted images, so it’s common in compositing applications and why we’ll call it the compositing-friendly direction.
If you need the calculation performed in the other direction, ie to REdistort with matchmoving-friendly distortion, or to UNdistort or matchmove with compositing-friendly distortion, it takes ~10 times longer to numerically invert the calculation, using numeric search techniques. Each computation gets performed on each pixel of an image, or many times when matchmoving. (You can’t use a solve from one direction in the other direction… but read on.)
Since there are trade-offs between the matchmoving-friendly and the compositing-friendly models, SynthEyes give you access to both kinds.
The compositing-friendly models are referred to in SynthEyes now as the “Standard” aka “Std.” models; the matchmoving-friendly models are referred to without a special prefix or indicator.
Note: Both Nuke and Fusion can work not only with compositing-friendly models, but also with matchmoving-friendly models; essentially the undistort and redistort nodes are reversed.
Here are suggestions as to how to decide which kind to use:
If you are delivering undistorted imagery using Lens Workflow #1, use the matchmoving-friendly versions.
If you are delivering distorted imagery, ie renders redistorted and composited over the original, as is usually the case with anamorphic footage, select the compositing-friendly “Standard” versions.
If you are using large lidar models, where slower mesh redraw time will add up over an extended tracking process in SynthEyes, consider using the matchmoving-friendly versions, then converting to the Standard version!
©2023 Boris FX, Inc. — UNOFFICIAL — Converted from original PDF.