It was cool to see my VR software being used by the Dallas Cowboys:
To make this a bit more of a technical post, instead of purely a brag, we use a fair amount of machine learning behind the scenes. We look for similar items in the scene between cameras, to match up and align the seams:
We try to ‘autotune’ all the parameters. We adjust both the seam position and the camera warp to minimize the squared RGB pixel difference in the overlapped areas. This is actually quite computationally expensive, so we run this in CUDA to accelerate it.