Use Recorded Camera Parameters (3D scene) to find closest 2D frames

Question

I am trying to replicate SCRC baseline model mentioned in the paper

The paper says “use the camera pose recorded with each annotation to extract nearest 2D frames for