FANet: Improving 3D Object Detection with Position Adaptation

Three-dimensional object detection plays a crucial role in achieving accurate and reliable autonomous driving systems. However, the current state-of-the-art two-stage detectors lack flexibility have limited feature extraction capabilities to effectively handle disorder irregularity of point clouds. In this paper, we propose novel network called FANet, which combines strengths PV-RCNN PAConv (position adaptive convolution). The goal FANet is address present our network, convolution operation constructs convolutional kernels using basic weight matrix, coefficients these are adaptively learned by LearnNet from relative points. This approach allows for flexible modeling complex spatial variations geometric structures 3D clouds, leading improved cloud features generation high-quality proposal boxes. Compared other methods, extensive experiments on KITTI dataset demonstrated that exhibits superior accuracy, showcasing significant improvement approach.