摘要 : Object detection on point clouds is widely used in autonomous driving technology. Recent studies have demonstrated that good feature representation is the key to 3-D object detection, especially for point-based methods. Meanwhile,... 展开 Object detection on point clouds is widely used in autonomous driving technology. Recent studies have demonstrated that good feature representation is the key to 3-D object detection, especially for point-based methods. Meanwhile, contrastive learning has shown its effectiveness in learning the general visual representation of images as a pretraining paradigm. Motivated by this, contrastive learning is extended to the task of object detection for better distinguishing different types of objects. In this article, a simple and effective single-stage detector, named 3-D object detection with balanced prediction based on contrastive point loss (BP-CPL), is proposed with a contrastive point loss (CPL) and a rebalanced branch (RBB). Through the CPL, similar representations among points of the same category and discriminative information among points of different categories can be learned. To keep the contrastive points consistent, a point filter is proposed. In addition, compared to the original point-wise feature, the self-supervised learned embedding is more robust to few-sample categories. Therefore, an RBB combined with the origin classification branch in a cumulative learning manner is proposed to rebalance the prediction results during the training phase. Extensive experiments show the effectiveness of the proposed method, especially for few-sample objects. The code will be available at https://github.com/Tongjiaxun/BP-CPL. 收起