Class incremental learning improves 3d object detection in autonomous driving
CL3DOD: Understanding the Stability-Plasticity Dilemma in Outdoor Class-Incremental 3D Object Detection
Computer Vision and Pattern Recognition
Summary
When teaching 3D object detectors to recognize new kinds of things on the road, they often forget what they learned before. The authors study this problem across different self-driving car datasets to see if it happens consistently. They created a method called CL3DOD that helps detectors remember old objects while learning new ones by using predictions from previous training stages and special knowledge-sharing techniques. Their method works better than others on multiple benchmarks, keeping a good balance between remembering and learning new stuff.
What this means in practice
- •For autonomous vehicle engineers: Enable continuous learning for self-driving cars to detect new object classes without losing accuracy on previously learned classes.
- •For robotics system developers: Improve incremental object detection models in mobile robots operating in changing outdoor environments by maintaining old knowledge while adapting to new objects.
Authors
Muhammad Ahmed Ullah Khan, Khalil Ahmed, Didier Stricker, Muhammad Zeshan Afzal
Abstract
Class-incremental learning (CIL) for outdoor 3D object detection remains largely underexplored. Existing outdoor studies are limited to single-benchmark evaluations, with SDCoT++ considering KITTI and CL2OD considering nuScenes under different incremental settings. This makes it difficult to assess whether observed continual-learning behavior generalizes across autonomous-driving benchmarks. To the best of our knowledge, we provide the first unified cross-benchmark analysis of outdoor class-incremental 3D object detection across multiple autonomous-driving benchmarks, covering both two-task and multi-stage settings. We investigate: (1) how severely existing 3D detectors forget previously learned classes, (2) how incremental training exposes the stability-plasticity dilemma between knowledge retention and new-class acquisition, and (3) which forms of supervision are most effective for balancing these objectives. We introduce CL3DOD, a class-incremental 3D detection framework that restores missing supervision for previous classes using predictions from the previous-stage detector, preserves old-class responses through query-wise knowledge distillation, and employs EMA-based temporal consistency regularization. Experiments on KITTI and nuScenes reveal a consistent stability-plasticity dilemma across both benchmarks. CL3DOD generally achieves a more favorable stability-plasticity balance than controlled incremental baselines and improves upon published prior methods in the reported benchmark comparisons. This advantage persists across multiple incremental stages, demonstrating strong retention of previously learned knowledge while maintaining effective new-class acquisition. Our code, evaluation setup, and incremental task splits will be publicly released.