AI summaryⓘ
The authors studied how different ways to handle imbalanced classes in 3D LiDAR data affect semantic segmentation, which helps computers understand scenes for navigation and mapping. They tested various methods on three datasets using two different algorithms, finding that some common techniques like inverse-frequency weighting can actually hurt performance, especially for rare classes. They also showed that the way data points are collected (structured vs. random sampling) and the type of data (real vs. synthetic) influence which methods work best. Overall, their work highlights that the choice of imbalance correction depends on sampling strategy, imbalance severity, and the nature of the data.
LiDARSemantic SegmentationClass Imbalance3D Point CloudsReweighting SchemesLoss FunctionsKPConvRandLA-NetStructured SamplingRandom Sampling
Authors
Antonis Savva, Christos Kyrkou, Theocharis Theocharides
Abstract
Class imbalance in LiDAR point clouds poses challenges for semantic segmentation in autonomous navigation and urban mapping. While 2D vision has numerous mitigation techniques, their effectiveness in 3D remains unclear. We benchmark six reweighting schemes and five imbalance-aware losses across three datasets (DALES, S3DIS, STPLS3D) using two architectures (KPConv, RandLA-Net). Inverse-frequency weighting degrades performance by up to 12% compared to uniform weighting, with catastrophic failures in minority classes. Uniform weighting performs within 2% of complex losses for structured sampling (KPConv) but benefits less for random sampling (RandLA-Net, up to 4.6% gap). Loss landscape analysis reveals a complex interplay: for structured sampling, imbalance ratio determines landscape geometry on real LiDAR data but decouples from it on synthetic data; for random sampling, landscapes show high sensitivity to dataset geometry regardless of imbalance ratio. For the two evaluated point-based architectures, these results suggest that the interaction between sampling strategy (structured vs. random), imbalance severity, and data acquisition characteristics shapes which mitigation approaches are effective.