SAGE3D: Soft-guided attention and graph excitation for 3D point cloud corner detection

2026-05-14

AI Summary

This study addresses the insufficient precision and recall in corner detection from airborne LiDAR point clouds by proposing a hierarchical encoder-decoder architecture that integrates Set Abstraction and Feature Propagation to enable multi-scale feature extraction and per-point prediction recovery. The core innovations include a soft-guided attention mechanism that incorporates ground-truth corner labels as log-prior information into the attention computation during training, and a unidirectional message-passing excitation graph neural network that amplifies high-confidence corner responses at critical network levels. Experimental results demonstrate that the proposed method effectively enhances corner signals, achieving significant improvements in both detection precision and recall across multiple datasets.

Abstract

We present SAGE3D, a hybrid Transformer-based model for corner detection in airborne LiDAR point clouds. We propose a multi-stage solution built on a hierarchical encoder-decoder architecture that progressively downsamples point clouds through Set Abstraction layers and recovers per-point predictions via Feature Propagation. We introduce two innovations: Soft-Guided Attention, which injects ground-truth corner labels as a log-prior into attention logits during training to improve precision; then an Excitatory Graph Neural Network positioned at strategic resolutions in the hierarchy, employing positive-only message passing where high-confidence corners reinforce predictions through learned boosting, optimizing for recall. The hierarchical design enables multi-scale feature extraction while our guided attention and excitatory modules ensure corner signals are amplified rather than diluted across scales.

Problem

• corner detection
• 3D point cloud
• LiDAR
• building reconstruction
• geospatial analysis

Innovation

• Soft-Guided Attention
• Excitatory Graph Neural Network
• 3D Point Cloud Corner Detection
• Hierarchical Encoder-Decoder
• Positive-only Message Passing