OctFormer: Octree-based Transformers for 3D Point Clouds

• Published in: ACM transactions on graphics Vol. 42; no. 4; pp. 1 - 11
• Main Author: Wang, Peng-Shuai
• Format: Journal Article
• Language: English
• Published: New York, NY, USA ACM 01.08.2023
• Subjects: Computer graphics; Computing methodologies; Machine learning; Machine learning approaches; Neural networks; Point-based models; Shape analysis; Shape modeling; point clouds; octree; transformers; 3D semantic segmentation; 3D object detection

We propose octree-based transformers, named OctFormer, for 3D point cloud learning. OctFormer can not only serve as a general and effective backbone for 3D point cloud segmentation and object detection but also have linear complexity and is scalable for large-scale point clouds. The key challenge in...