Authors: Kavisha Vidanapathirana, Shin-Fang Chng, Xueqian Li, Simon Lucey

Published on: October 16, 2023

Impact Score: 8.07

Arxiv code: Arxiv:2310.10301

Summary

• What is new: Introduced a method for achieving multi-body rigidity in scene flow without constraining SE(3) parameters, utilizing isometry to regularize flow predictions.
• Why this is important: Existing approaches struggle to identify multi-body rigid motions in real-world data and rely on cumbersome, brittle strategies.
• What the research proposes: A novel regularization technique that encourages isometry in flow predictions for rigid bodies, enabling multi-body rigidity while maintaining a continuous flow field.
• Results: Outperformed state-of-the-art in 3D scene flow and long-term point-wise 4D trajectory prediction on real-world datasets.

Technical Details

Technological frameworks used: Test-time optimization using coordinate networks

Models used: Isometry-based regularization for scene flow

Data used: Real-world datasets

Potential Impact

Automotive industry, robotics, AR/VR companies, and companies involved in autonomous vehicle development.

Want to implement this idea in a business?

We have generated a startup concept here: FlowVision.