This is a somewhat refactored PIPs with a batched chained prediction. It provides the setup to evaluate PIPs on the TAP-Vid dataset. Also, a basic integration of the point trajectories from PIPs (or RAFT) with Sam is provided, to get video segmentation.