激光雷达数据由于其几何特性，被广泛应用于三维目标检测任务中。由于点云数据的稀疏性和不规则性，难以实现特征提取的质量和推理速度间的平衡。本文提出一种基于体柱特征编码的三维目标检测算法，以Pointpillars网络为基础，设计Teacher-Student模型框架对回归框尺度进行蒸馏，增加蒸馏损失，优化训练网络模型，提升特征提取的质量。为进一步提高模型检测效果，设计定位引导分类项，增加分类预测和回归预测之间的相关性，提高物体识别准确率。本网络所做改进没有引入额外的网络嵌入。算法在KITTI数据集上的实验结果表明，相比于基准网络，在三维模式下的平均精度值从60.65%提升到了64.69%，鸟瞰图模式下的平均精度值从67.74%提升到70.24%。模型推理速度为45 FPS，在提升检测精度的同时满足了实时性要求。

Developing low-cost, efficient, and stable photocatalysts is one of the most promising methods for large-scale solar water splitting. As a metal-free semiconductor material with suitable band gap, graphitic carbon nitride (g-C₃N₄) has attracted attention in the field of photocatalysis, which is mainly attributed to its fascinating physicochemical and photoelectronic properties. However, several inherent limitations and shortcomings—involving high recombination rate of photocarriers, insufficient reaction kinetics, and optical absorption—impede the practical applicability of g-C₃N₄. As an effective strategy, vacancy defect engineering has been widely used for breaking through the current limitations, considering its ability to optimize the electronic structure and surface morphology of g-C₃N₄ to obtain the desired photocatalytic activity. This review summarizes the recent progress of vacancy defect engineered g-C₃N₄ for solar water splitting. The fundamentals of solar water splitting with g-C₃N₄ are discussed first. We then focus on the fabrication strategies and effect of vacancy generated in g-C₃N₄. The advances of vacancy-modified g-C₃N₄ photocatalysts toward solar water splitting are discussed next. Finally, the current challenges and future opportunities of vacancy-modified g-C₃N₄ are summarized. This review aims to provide a theoretical basis and guidance for future research on the design and development of highly efficient defective g-C₃N₄.