EUVIMEDEUVIMED entdecken
Crossref

SCVM3D: A Hybrid Backbone With Sparse Convolutions and Mamba-2 for Voxel-Based 3D Object Detection

Lei Zhang, Zhaozhong Wang, Yongqiang Zhao, Bin Fan, Nian Liu, Binglu Wang

IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 16100-16113

DOI: 10.1109/tcsvt.2026.3712403 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Lei Zhang und 5 weiteren Autor:innen. Die Arbeit nennt 93 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate and efficient 3D object detection is crucial for autonomous driving, where LiDAR-based methods are essential for capturing high-precision spatial information in large-scale outdoor environments. Among existing approaches, voxel-based methods, which partition point clouds into structured grids, offer a computationally efficient paradigm. However, these approaches struggle to model long-range dependencies and capture global contextual information. Recent advances in state-space models (SSMs), particularly the Mamba architecture, offer a promising solution by enabling efficient global feature modeling with linear complexity. Nevertheless, the inherently sequential nature of SSMs is misaligned with the spatially unordered structure of voxel data. In this paper, we propose SCVM3D, a novel backbone architecture that combines sparse convolution and the SSM-based Mamba-2 model to address the limitations of voxel-based 3D object detection. SCVM3D integrates three core components: 1) a hierarchical sparse convolution module that efficiently aggregates local geometric cues, 2) an importance-aware adaptive voxel reordering scheme that optimizes the token sequence, and 3) a Mamba-2 module that captures global contextual relationships across the scene. By dynamically prioritizing voxel features and using the linear-complexity modeling capability of Mamba-2 for long-range dependencies, SCVM3D effectively captures both local and global information from voxelized LiDAR data. Extensive experiments on both the KITTI and nuScenes datasets demonstrate that SCVM3D consistently improves the performance of strong baseline detectors across diverse environments. These results underscore the potential of SCVM3D as an effective and efficient backbone for large-scale 3D object detection.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Zhang, L., Wang, Z., Zhao, Y., Fan, B., Liu, N., & Wang, B. (2026). SCVM3D: A Hybrid Backbone With Sparse Convolutions and Mamba-2 for Voxel-Based 3D Object Detection. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 16100-16113. https://doi.org/10.1109/tcsvt.2026.3712403

Vancouver

Zhang L, Wang Z, Zhao Y, Fan B, Liu N, Wang B. SCVM3D: A Hybrid Backbone With Sparse Convolutions and Mamba-2 for Voxel-Based 3D Object Detection. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):16100-16113. doi:10.1109/tcsvt.2026.3712403

BibTeX
@article{Zhang2026,
  author = {Zhang, Lei and Wang, Zhaozhong and Zhao, Yongqiang and Fan, Bin and Liu, Nian and Wang, Binglu},
  title = {SCVM3D: A Hybrid Backbone With Sparse Convolutions and Mamba-2 for Voxel-Based 3D Object Detection},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {11},
  pages = {16100--16113},
  doi = {10.1109/tcsvt.2026.3712403}
}
Kostenlos starten →