Presentation / Installation


OctFormer: Octree-based Transformers for 3D Point Clouds
SessionDeep Geometric Learning
DescriptionOctFormer is a general, effective, and scalable backbone for 3D point cloud learning. Octformer leverages a novel octree attention that partitions point clouds into irregular local windows containing a fixed number of points. OctFormer achieves state-of-the-art performances on 3D segmentation on ScanNet/ScanNet200 and 3D detection on SUN RGB-D.
Author
Peng-Shuai Wang
Peking University
Event Type
Technical Paper
TimeThursday, 10 August 20232pm - 2:11pm PDT
LocationRoom 515 A
ACM Digital Library Technical Papers pdfs
Session Time & Location
Sunday, 6 August 20236pm - 8:30pm PDTWest Hall B
Thursday, 10 August 20232pm - 3:30pm PDTRoom 515 A
Interest Areas
Research & Education
Recordings
Livestreamed
Recorded
Keywords
Modeling
Rendering
Registration Categories
FC
FCS
V
VS
EFC
Next PresentationNext Presentation
Dictionary Fields: Learning a Neural Basis Decomposition