Abstract:The semantic segmentation network performance of large-scale outdoor point clouds is usually limited by the number of input point clouds. In the application of most methods, the point cloud is cut into small pieces as the training input, which will not only lead to heavy preprocessing burdens but also destroy the overall geometric structure of the scene. The transformer network has demonstrated remarkable advantages of attention mechanisms in focusing on crucial features and improving model performance. However, its training and inference on large-scale data are confined by computational complexity. To eliminate these challenges, the attention mechanisms are improved to enhance their performance in processing large-scale input data, while removing limitations imposed by computational complexity. Moreover, a novel Local Feature Enhancement (LFE) module is developed to construct the LFE-Net, which can accurately and efficiently extract spatial and attribute features from local point clouds. In particular, an improved attention module, which is called Simple Offset Attention (SOA), is adopted for local point cloud spatial feature learning. Compared with self-attention, SOA requires less memory and can better capture the fine-grained local features. Furthermore, to effectively avoid the destruction of object geometry and diminish the impact of sample imbalance, a training sample collection method based on the number of different classes is designed. To validate the effectiveness of this method, some experiments are conducted based on publicly accessible outdoor point cloud datasets. The results demonstrate that the LFE-Net can achieve substantial improvements compared with other cutting-edge network models.

Separable Self-Attention Mechanism for Point Cloud Local and Global Feature Modeling

Point Attention Network for Point Cloud Semantic Segmentation.

Attention-based Transformation from Latent Features to Point Clouds

Local Transformer Network on 3D Point Cloud Semantic Segmentation

Enhanced Local Feature Learning with Simple Offset Attention for Semantic Segmentation of Large-scale Point Clouds

Associate Semantic-Instance Segmentation of 3D Point Clouds Based on Local Feature Extraction

Point Cloud Classification Network Based on Self-Attention Mechanism.

Rethinking Local-to-global Representation Learning for Rotation-Invariant Point Cloud Analysis

CloudAttention: Efficient Multi-Scale Attention Scheme For 3D Point Cloud Learning

DSACNN: Dynamically Local Self-Attention CNN for 3D Point Cloud Analysis.

MPCT: Multiscale Point Cloud Transformer with a Residual Network

Bi-Directional Attention for Joint Instance and Semantic Segmentation in Point Clouds

Cyclic Self-attention for Point Cloud Recognition.

Point Attention Network for Semantic Segmentation of 3D Point Clouds

Global Attention-Guided Dual-Domain Point Cloud Feature Learning for Classification and Segmentation

Attention-Based Deep Neural Network for Point Cloud Learning

Multi-Head Self-Attention for 3D Point Cloud Classification

Rethinking Attention Module Design for Point Cloud Analysis

Modify Self-Attention Via Skeleton Decomposition for Effective Point Cloud Transformer.

Point Gated Attention

3DPCT: 3D Point Cloud Transformer with Dual Self-attention