prmanTracker produces fast and accurate 2d tracking data to match any surface of images rendered using Pixar's Renderman.

Please note this will only work when using it on exr images rendered with Pixar's Renderman that includes a worldToNDC matrix in their metadata.

Here's how it works:

• 2 inputs. Image input + Position input (this is a position or pworld pass rendered out of prman)
• Connect the relevant inputs (see if you need to change the positionLayer, which is the layer that stores position data in your pos input)
• Move the 4 points in the viewer over features in your image you'd like to get tracks for (just like you would on a regulartracker). (you can deactivate points if you don't need all 4)
• Click "Track it!"
• Confirm the frame range to be tracked, and wait for it to finish.

The process should be pretty fast, and it's a lot faster if you repeat the same process with the same image.

What it actually does is take a sample of the position pass at each one of the enabled points to get a 3d coordinate. Then uses the camera matrix in the exr metadata to do an inverse projection of that point onto screen coordinates, repeats and sets a keyframe for each frame.
When you're done, you can choose to get those points into a new Tracker node, or a Cornerpin node if you have all 4 tracks enabled.

Limitations: Since it samples from the actual image, you can only track points that "exist" in the image. So, you can track any surface you can see, but not a random position in mid air. However, you only need to "see" the surface you want to track for one frame. If you're tracking a surface (say a table), and the table goes off frame, you'll still get trackers for it, because you still have camera data in the exr.

If you want to get 2d tracks for arbitrary points in 3d space, use prmanReconcile instead (to be uploaded soon).