Monocular Open Vocabulary Occupancy Prediction for Indoor Scenes (LegoOcc)

LegoOcc tackles monocular open-vocabulary 3D semantic occupancy in large-scale indoor scenes under geometry-only supervision (binary occupancy labels only). We represent scenes as Language-Embedded Gaussians, propose an opacity-aware Poisson Gaussian-to-Occupancy operator for stable volumetric aggregation, and introduce Progressive Temperature Decay to reduce feature mixing during splatting and strengthen Gaussian–language alignment.

---

News

• Accepted to CVPR2026, code will be released before the conference

---

Citation

If you find this work useful, please consider citing:

@misc{zhou2026monocularopenvocabularyoccupancy,
      title={Monocular Open Vocabulary Occupancy Prediction for Indoor Scenes}, 
      author={Changqing Zhou and Yueru Luo and Han Zhang and Zeyu Jiang and Changhao Chen},
      year={2026},
      eprint={2602.22667},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2602.22667},
}