PointViG: A Lightweight GNN-based Model for Efficient Point Cloud Analysis

Qiang Zheng,Yafei Qi,Chen Wang,Chao Zhang,Jian Sun

2024-09-16

Abstract:In the domain of point cloud analysis, despite the significant capabilities of Graph Neural Networks (GNNs) in managing complex 3D datasets, existing approaches encounter challenges like high computational costs and scalability issues with extensive scenarios. These limitations restrict the practical deployment of GNNs, notably in resource-constrained environments. To address these issues, this study introduce <b>Point<\b> <b>Vi<\b>sion <b>G<\b>NN (PointViG), an efficient framework for point cloud analysis. PointViG incorporates a lightweight graph convolutional module to efficiently aggregate local features and mitigate over-smoothing. For large-scale point cloud scenes, we propose an adaptive dilated graph convolution technique that searches for sparse neighboring nodes within a dilated neighborhood based on semantic correlation, thereby expanding the receptive field and ensuring computational efficiency. Experiments demonstrate that PointViG achieves performance comparable to state-of-the-art models while balancing performance and complexity. On the ModelNet40 classification task, PointViG achieved 94.3% accuracy with 1.5M parameters. For the S3DIS segmentation task, it achieved an mIoU of 71.7% with 5.3M parameters. These results underscore the potential and efficiency of PointViG in point cloud analysis.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to solve the problems of high computational cost and scalability in large - scale scenarios faced by Graph Neural Networks (GNNs) in point - cloud analysis. Specifically, although existing methods perform well in handling complex 3D datasets, their computational complexity is high, and it is especially difficult to achieve efficient deployment in resource - constrained environments. These problems limit the wide use of GNNs in practical applications.

To address these challenges, this paper proposes a lightweight framework named Point VisionGNN (PointViG), aiming to improve the efficiency of point - cloud analysis. The main contributions of PointViG include:

1. **Lightweight Graph Convolution Module**: This module effectively compresses the computational process by optimizing the operations on neighbor - node tensors, significantly reducing the overall computational load. In addition, the module design includes strategies to enhance feature diversity to mitigate the over - smoothing phenomenon, thereby improving the feature representation ability.

2. **Adaptive Dilated Graph Convolution Strategy**: This strategy is optimized for large - scale point - cloud scenarios. By searching for sparse neighbor - nodes of each node in the low - dimensional geometric space and adaptively selecting neighbor - nodes based on high - dimensional semantic associations, it effectively expands the receptive field while maintaining a low computational cost. This method not only enlarges the perception range of each node but also ensures the retention of key semantic information during the sampling process.

3. **Experimental Verification**: Through extensive experiments, PointViG has demonstrated performance comparable to existing state - of - the - art methods on multiple tasks while maintaining a better balance in the number of parameters and computational complexity. For example, in the ModelNet40 classification task, PointViG achieved an accuracy of 94.3% with only 1.5M parameters; in the S3DIS segmentation task, it achieved an mIoU of 71.7% with 5.3M parameters.

In conclusion, through its lightweight graph convolution module and adaptive dilated graph convolution strategy, PointViG not only improves computational efficiency but also enhances the processing ability for large - scale point clouds, providing an efficient solution for point - cloud analysis.