Official datasets and checkpoints of the paper -- Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling