Authors: Damien Robert, Hugo Raguet, Loic Landrieu

Published on: January 12, 2024

Impact Score: 8.22

Arxiv code: Arxiv:2401.06704

Summary

• What is new: A highly efficient method for panoptic segmentation of large 3D point clouds using scalable graph clustering, eliminating the need for instance-matching during training.
• Why this is important: The resource-intensive instance-matching step in the panoptic segmentation of large 3D point clouds.
• What the research proposes: Redefining the task as a scalable graph clustering problem that can be adapted to the superpoint paradigm, making it more efficient.
• Results: Achieved state-of-the-art performance on indoor scanning datasets S3DIS Area 5 and ScanNetV2, as well as set the first state-of-the-art for KITTI-360 and DALES, with a significantly smaller and faster model.

Technical Details

Technological frameworks used: Graph Clustering

Models used: SuperCluster

Data used: S3DIS Area 5, ScanNetV2, KITTI-360, DALES

Potential Impact

3D mapping, autonomous driving, real estate and construction industries, VR and AR development companies.

Want to implement this idea in a business?

We have generated a startup concept here: SuperMapping.