CL3D: Camera-LiDAR 3D Object Detection With Point Feature Enhancement and Point-Guided Fusion

Camera-LiDAR 3D object detection has been extensively investigated due to its significance for many real-world applications. However, there are still of great challenges address the intrinsic data difference and perform accurate feature fusion among two modalities. To these ends, we propose a two-stream architecture termed as CL3D, that integrates with point enhancement module, point-guided module IoU-aware head cross-modal detection. Specifically, pseudo LiDAR is firstly generated from RGB image, (PEM) then designed enhance raw point. Moreover, (PFM) developed find image-point correspondence at different resolutions, incorporate semantic geometric features in point-wise manner. We also investigate inconsistency between localization confidence classification score detection, introduce prediction (IoU Head) box regression. Comprehensive experiments conducted on publicly available KITTI dataset, CL3D reports outstanding performance compared both single- multi-modal detectors, demonstrating effectiveness competitiveness.