Tsukada Lab. Computer Network Group

SUG-Occ: Explicit Semantics and Uncertainty Guided Sparse Learning for Efficient 3D Occupancy Prediction

Hanlin Wu, Pengfei Lin, Ehsan Javanmardi, Naren Bao, Bo Qian, Hao Si, Manabu Tsukada

, Jan 16, 2026

Summary in three lines

  • SUG-Occ makes 3D semantic occupancy prediction more efficient by exploiting the sparsity of real scenes.
  • Semantic and depth uncertainty guide which voxels are kept, and sparse completion recovers finer structure.
  • Evaluation on SemanticKITTI reports gains in occupancy accuracy and computational efficiency.

Figures from the paper

Figure 2: SUG-Occ pipeline
Figure 2. SUG-Occ pipelineSource: Figure 2 from Wu et al., arXiv (2026) (authors’ figure from the manuscript)Semantic and depth uncertainty guide sparse projection before occupancy completion.

How to cite

H. Wu, P. Lin, E. Javanmardi, N. Bao, B. Qian, H. Si, and M. Tsukada, “SUG-Occ: Explicit Semantics and Uncertainty Guided Sparse Learning for Efficient 3D Occupancy Prediction,” in , 2026.

BibTeX

@unpublished{Wu2026a,
  title         = {SUG-Occ: Explicit Semantics and Uncertainty Guided Sparse Learning for Efficient 3D Occupancy Prediction},
  author        = {Hanlin Wu and Pengfei Lin and Ehsan Javanmardi and Naren Bao and Bo Qian and Hao Si and Manabu Tsukada},
  year          = {2026},
  date          = {2026-01-16},
  eprint        = {2601.11396},
  archiveprefix = {arXiv},
  primaryclass  = {cs.CV},
  url           = {https://arxiv.org/abs/2601.11396},
  note          = {Preprint}
}