基于多尺度特征融合和对比池化的点云补全网络

点云补全在点云处理任务中具有重要作用，它可以提高数据质量、辅助生成精确三维模型，为多种应用提供可靠数据支撑。然而，现有基于深度网络的点云补全算法采用的单层次全局特征提取方法较为简单，没有充分挖掘潜在语义信息，并在编码过程中丢失部分细节信息。为解决这些问题，提出了一种多尺度特征逐级融合的点云补全网络，并结合注意力机制提出了一种全新的池化方法。实验结果表明，在PCN、ShapeNet34和ShapeNet55三个数据集上取得了SOTA水平，证明该网络具有更好的特征表示能力和补全效果。

Multi-scale feature fusion and contrastive pooling for point cloud completion network

Point cloud completion plays a crucial role in point cloud processing tasks, as it enhances data quality, assists in generating accurate 3D models, and provides reliable data support for various applications. However, existing point cloud completion algorithms based on deep neural networks use a simple single-level global feature extraction method, which do not fully exploit latent semantic information and lead to loss some detailed information during the encoding process. To address these issues, this paper proposed a novel point cloud completion network that employed a multi-scale feature fusion approach and introduced a new pooling method by combining an attention mechanism. Experimental results demonstrate that the proposed network achieves the state-of-the-art(SOTA) performance on three datasets, namely PCN, ShapeNet34, and ShapeNet55, indicating its superior feature representation capability and completion effectiveness.