EPDet: Enhancing point clouds features with effective representation for 3D object detection

Yidong Chen, Guorong Cai, Qiming Xia, Zhaoliang Liu, Binghui Zeng, Zongliang Zhang, Jonathan Li, Zongyue Wang

Abstract: Highlights•We propose the BEV Offset Transformer, which is a plug-and-play module, and its robustness is validated through numerous experiments.•We conduct experiments to fill the gap in multi-modal fusion for Focal Conv and propose a more effective Pyramid-like Convolution approach.•EPDet (CasA-based) achieves satisfactory results on the KITTI dataset.