Cross-Modality Knowledge Distillation Network for Monocular 3D Object Detection
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Leveraging LiDAR-based detectors or real LiDAR point data to guide monocular 3D detection has brought significant improvement, e.g., Pseudo-LiDAR methods. However, the existing methods usually apply non-end-to-end training strategies and insufficiently leverage the LiDAR information, where the rich potential of the LiDAR data has not been well exploited. In this paper, we propose the Cross-Modality Knowledge Distillation (CMKD) network for monocular 3D detection to efficiently and directly transfer the knowledge from LiDAR modality to image modality on both features and responses. Moreover, we further extend CMKD as a semi-supervised training framework by distilling knowledge from large-scale unlabeled data and significantly boost the performance. Until submission, CMKD ranks 1 st among the monocular 3D detectors with publications on both KITTI test set and Waymo val set with significant performance gains compared to previous state-of-the-art methods. Our code will be released at https://github.com/Cc-Hy/CMKD.
Object detection, Optical radar
Y. Hong, H. Dai, and Y. Ding. Cross-Modality Knowledge Distillation Network for Monocular 3D Object Detection, in Computer Vision (ECCV 2022), Lecture Notes in Computer Science, Oct 2022, vol 13670, pp. 87-104, doi:10.1007/978-3-031-20080-9_6
IR conditions: non-described