Checkpoints of paper: LRVOS: Learning to Reason and Segment over Long-term Videos