LoDA's new 3D change-detection pipeline hits 95% accuracy, beats baselines

A team of researchers led by Haitian Wang has published LoDA, a novel object-level 3D change-detection pipeline and benchmark designed for high-definition LiDAR maps used in autonomous driving and smart-city services. Unlike existing tile-based, threshold-driven approaches that produce per-point scores without explicit detection limits, LoDA decouples registration, geometry, and semantics to propagate pose uncertainty into spatially varying detection limits. It uses geometry-driven object proxies with rule-based semantic and instance segmentation, plus displacement cues in height, volume, and surface-normal direction to assign five change labels with confidence. This suppresses false changes caused by residual misalignment and density variation across multi-temporal urban LiDAR scans.

To validate the method, the authors built the LoDA benchmark for the Subiaco district, fusing multi-temporal vehicle-LiDAR maps with GNSS and IMU support, semantic instances, and object-level annotations. On this benchmark, LoDA achieves 95.0% accuracy, 90.8% macro F1, and 83.0% macro IoU, exceeding the best baseline by 8.7 IoU points and 4.4 F1 points. On the public Urb3DCD-V2 benchmark under the official point-wise protocol, it reaches 96.81% mean accuracy and 89.52% mean change IoU, improving on the strongest reported baselines by 1.36 mAcc and 3.18 mIoU. The work is accepted at the 34th ACM International Conference on Multimedia (ACM MM 2026) and is available on arXiv (2608.05356).