Intrinsic-feature-guided 3D object detection

LiDAR-based 3D object detection is essential for autonomous driving systems. However, LiDAR point clouds usually exhibit sparsity, uneven distribution, and incomplete structures. In road driving environments, target objects within a class, such as vehicles, pedestrians, and cyclists, contain common intrinsic structural features thus, these objects can be well-suited for enhancing representation through the guidance of complete templates. Therefore, this paper presents an intrinsic-feature-guided 3D object detection method based on a template-assisted feature enhancement module, which extracts intrinsic features from relatively generalized templates and provides rich structural information for foreground objects. Furthermore, a proposal-level supervised contrastive learning mechanism is designed to enhance the feature differences between foreground and background objects, enabling the model to distinguish object classes better. The proposed modules are plug-and-play, which can always improve the performance of existing methods. Extensive experiments illustrate that the proposed method achieves highly competitive detection results. Code will be available at https://github.com/zhangwanjingjj/IfgNet.git.