PI3DETR: Parametric Instance Detection of 3D Point Cloud Edges with a Geometry-Aware 3DETR

2025-09-03

AI Summary

This work addresses the challenge of detecting parametric 3D curves directly from raw LiDAR point clouds in realistic 3D perception scenarios. We propose the first end-to-end method that regresses multiple types of 3D parametric curve instances—such as lines, circles, and ellipses—directly from point clouds, eliminating intermediate representations and multi-stage pipelines. Built upon the 3DETR architecture, our approach introduces a geometry-aware attention mechanism, a curve-specific bipartite matching strategy and loss function, and a lightweight post-processing module. Our key contribution is the first single-forward-pass framework enabling joint detection of diverse curve types with robust geometric parameter estimation. Evaluated on the ABC dataset, our method achieves state-of-the-art performance and demonstrates strong generalization and robustness to real-world sensor noise and variable point density.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Learning with Manifolds
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work

Abstract

We present PI3DETR, an end-to-end framework that directly predicts 3D parametric curve instances from raw point clouds, avoiding the intermediate representations and multi-stage processing common in prior work. Extending 3DETR, our model introduces a geometry-aware matching strategy and specialized loss functions that enable unified detection of differently parameterized curve types, including cubic Bézier curves, line segments, circles, and arcs, in a single forward pass. Optional post-processing steps further refine predictions without adding complexity. This streamlined design improves robustness to noise and varying sampling densities, addressing critical challenges in real world LiDAR and 3D sensing scenarios. PI3DETR sets a new state-of-the-art on the ABC dataset and generalizes effectively to real sensor data, offering a simple yet powerful solution for 3D edge and curve estimation.

Problem

• Detecting 3D parametric curve instances from raw point clouds
• Unifying detection of multiple curve types in single pass
• Improving robustness to noise and varying sampling densities

Innovation

• End-to-end parametric curve prediction from point clouds
• Geometry-aware matching and specialized loss functions
• Unified detection of multiple curve types in single pass