How to get ground truth disparity at inference time

[cliffsong94]

Hi author,
I am confused about how to get ground truth disparity at forward diffusion stage in inference time. I am trying to think of using this idea to refine my autonomous driving stereo application (I have a stereo camera rig and get disparity at a specific timestamp by using two images), and it looks like there is no way I can get gt disparity. I am confused about this idea or may need clarification as in the real app there is no way of having gt disparity.

Thanks