HITshan/DK-VLM-RCdata

Reference-Frame Composition (RC)

Paper

RC provides stepwise supervision for reference-frame updates in multi-step spatial reasoning, addressing a supervision gap in single-step spatial question answering.

This repository provides approximately 1.6M RC annotations with 2–4 steps, constructed from verified single-step annotations in VSI-590K and SPAR. The paper uses 150K filtered and sampled two- and three-step examples for training.

Original images and videos are not included; obtain them from the corresponding source datasets. See GitHub for usage instructions.

Citation

@misc{su2026dkvlmgeometrypositionspatial,
  title={DK-VLM: Geometry as Position for Spatial Reasoning in Vision-Language Models},
  author={Han Su and Meng Tian and Tianyu Huang and Mingyu Gao and Lennox Wei and Dongxu Li and Jianhua Han and Hang Xu and Xiaohe Wu and Wangmeng Zuo},
  year={2026},
  eprint={2610.22356},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2610.22356},
}