HIMap: HybrId Representation Learning for End-to-end Vectorized HD Map Construction This paper proposes a simple yet effective HybrId framework named HIMap for end-to-end vectorized HD map construction. HIMap introduces a hybrid representation called HIQuery to represent all map elements and a point-element interactor to interactively extract and encode both point-level and element-level information. Additionally, a point-element consistency constraint is proposed to enhance the consistency between point-level and element-level information. The output point-element integrated HIQuery can be directly converted into map elements' class, point coordinates, and mask. Extensive experiments show that HIMap consistently outperforms previous methods on both nuScenes and Argoverse2 datasets, achieving 77.8 mAP on nuScenes and 72.7 mAP on Argoverse2. The method achieves new state-of-the-art results by effectively learning and interacting both point-level and element-level information, leading to more accurate and complete map elements. The framework is evaluated on both nuScenes and Argoverse2 datasets, and the results show that HIMap outperforms previous methods in terms of mAP. The method is also extended to 3D map construction and centerline learning, achieving further improvements. The supplementary material provides additional analysis of the proposed HIMap, including more implementation details, inference speed, memory, and model size, as well as more ablation studies and qualitative analysis.HIMap: HybrId Representation Learning for End-to-end Vectorized HD Map Construction This paper proposes a simple yet effective HybrId framework named HIMap for end-to-end vectorized HD map construction. HIMap introduces a hybrid representation called HIQuery to represent all map elements and a point-element interactor to interactively extract and encode both point-level and element-level information. Additionally, a point-element consistency constraint is proposed to enhance the consistency between point-level and element-level information. The output point-element integrated HIQuery can be directly converted into map elements' class, point coordinates, and mask. Extensive experiments show that HIMap consistently outperforms previous methods on both nuScenes and Argoverse2 datasets, achieving 77.8 mAP on nuScenes and 72.7 mAP on Argoverse2. The method achieves new state-of-the-art results by effectively learning and interacting both point-level and element-level information, leading to more accurate and complete map elements. The framework is evaluated on both nuScenes and Argoverse2 datasets, and the results show that HIMap outperforms previous methods in terms of mAP. The method is also extended to 3D map construction and centerline learning, achieving further improvements. The supplementary material provides additional analysis of the proposed HIMap, including more implementation details, inference speed, memory, and model size, as well as more ablation studies and qualitative analysis.