Coarse to fine-based image-point cloud fusion network for 3D object detection

Meilan Hao, Zhongkang Zhang, Lei Li, Kejian Dong, Long Cheng, Prayag Tiwari, Xin Ning

Abstract: Highlights•Quantized perception strategy finely maps image pixels to point clouds, enhancing 3D virtual point clouds.•Introduced dynamic 2D Gaussian distribution to refine feature matching accuracy.•Implemented a density-aware RoI network to filter and refine detection features.