CSOT: Cross-Scan Object Transfer for Semi-Supervised LiDAR Object Detection

Jinglin Zhan, Tiejun Liu, Rengang Li, Zhaoxiang Zhang, Yuntao Chen* ;

"Large-scale 3D bounding box annotation is crucial for LiDAR object detection but comes at a high cost. Semi-supervised object detection (SSOD) offers promising solutions to leverage unannotated data, but the predominant pseudo-labeling approach requires careful hyperparameter tuning for training on noisy teacher labels. In this work, we propose a () paradigm for LiDAR SSOD. Central to our approach is , a transformer-based network that predicts possible placement locations and the object-place fitness scores for inserting annotated objects into unlabeled scans in a semantic coherence manner. Based on , successfully enables object copy-paste in LiDAR SSOD for the first time. To train object detectors on partially annotated scans generated by , we adopt a spatial-aware classification loss throughout our partial supervision to handle false negative issues caused by treating all unlabeled objects as background. We conduct extensive experiments to verify the efficacy and generality of our method. Compared to other state-of-the-art label-efficient methods used in LiDAR detection, our approach requires the least amount of annotation while achieves the best detector. Using only 1% of the labeled data on the Waymo dataset, our semi-supervised detector achieves performance on par with the fully supervised baseline. Similarly, on the nuScenes dataset, our semi-supervised CenterPoint reaches 99% of the fully supervised model’s detection performance in terms of NDS score, while using just 5% of the labeled data. Code is released at https://github.com/JinglinZhan/CSOT"