Geometry improves sampling for rare objects in 3D LiDAR detection
GA-EIRFS: A Geometry-Augmented Repeat-Factor Sampling Method for Long-Tailed LiDAR 3D Object Detection
Computer Vision and Pattern Recognition
Summary
Detecting different types of objects in 3D LiDAR data can be difficult when some classes appear less often. The authors found that just counting how often an object appears isn't enough because some objects have more geometric details visible in LiDAR scans than others. They created a method that adjusts how often certain frames are used during training by considering both object frequency and geometric information like point density and surface shape. This change helps detection models better recognize rare objects without changing the models themselves or how they make predictions.
What this means in practice
- •For autonomous vehicle engineers: Improve detection accuracy of rare object classes in LiDAR-based perception systems by better sampling training data using geometric cues.
- •For robotics perception teams: Enhance the training process of 3D detectors to recognize less frequent but geometrically distinctive objects by modifying data sampling without changing model architecture.
Authors
Taufiq Ahmed, Constantino Álvarez Casado, Daniel Herrera Castro, Sasan Sharifipour, Abhishek Kumar, Miguel Bordallo López
Abstract
Long-tailed 3D object detection is treated as a class-frequency problem, but LiDAR supervision quality depends on object observability: similar frequencies can hide different geometric evidence. We introduce Geometry-Augmented Exponentially Weighted Instance-Aware Repeat Factor Sampling (GA-EIRFS), a detector-agnostic method that modulates a frequency-based repeat factor with a fixed geometry score combining point count, surface-normal entropy, and surface coverage. GA-EIRFS changes only frame-sampling probabilities, leaving the detector and inference unchanged. On nuScenes it improves mean average precision (mAP) and the nuScenes detection score (NDS) in four converged experiments with CenterPoint and PointPillars over two seeds; for CenterPoint at seed 666, mAP rises from 0.552 to 0.563 and bicycle AP from 0.306 to 0.359. Per-class gains correlate with the class sampling-weight increase (Spearman rho=0.70, p=0.025) but not with geometry score alone (rho=0.32, p=0.37), so geometry amplifies frequency-driven need. KITTI results vary across seeds, most for the rarest class. Code: https://github.com/Multimodal-Sensing-Lab/GA-EIRFS.