Multi-view fusion pillar-based point clouds object detection network

What is it about?

we present a novel and efficient two-stage point-pillar hybrid architecture named Attentive Multi-View Fusion Network (AMVFNet), in which we abstract features from all cylindrical view, bird-eye view, and raw point clouds. Rather than designing more complex modules to solve the problems inherent in the single-view approach, our multi-view fusion architecture effectively combines the strengths of multiple perspectives to improve performance at a more fundamental level. Besides, to compensate for quantization distortion caused by projection operations, we propose attentive feature enhancement layers to further improve the capability of contextual information capturing.

Featured Image

Why is it important?

1) We present a novel two-stage 3D object detection network, namely AMVFNet, which extracts objective features from all CYV and BEV along with raw point clouds and makes full use of complementary information in the form of these data representations. 2) A multi-view feature fusion module based on the cross-view feature gating mechanism is proposed to guide the complementarity and integration of features between different viewpoints. 3) We introduce two different feature enhancement modules around attention mechanism proposed in the first and the second stage, which improve model’s capability to characterize geometrical and contextual information.

Perspectives