MLGCN: an ultra efficient graph convolutional neural model for 3D point cloud analysis.

Area of Science:

• Computer Vision
• Machine Learning
• 3D Data Analysis

Background:

• 3D sensors (LiDAR, scanners, RGB-D cameras) are increasingly accessible, generating large point cloud datasets.
• Deep learning models for 3D point cloud analysis often have high computational costs, limiting real-time applications.
• Efficient algorithms are crucial for 3D model classification and segmentation.

Purpose of the Study:

• To propose an ultra-efficient deep learning model for 3D point cloud analysis.
• To reduce the computational overhead and memory usage of existing 3D analysis models.
• To enable real-time 3D analysis on resource-constrained devices.

Main Methods:

• Developed the Multi-level Graph Convolutional Neural Network (MLGCN), a lightweight, graph-based architecture.
• Utilized shallow Graph Neural Network (GNN) blocks to extract features at multiple spatial locality levels.
• Leveraged precomputed k-Nearest Neighbors (KNN) graphs shared across GCN blocks to minimize computation.

Main Results:

• MLGCN achieves competitive performance in 3D object classification and part segmentation.
• The model requires up to 1000x fewer floating-point operations and significantly less storage compared to state-of-the-art models.
• Demonstrated suitability for deployment on low-memory and low-CPU devices.

Conclusions:

• MLGCN offers a highly efficient solution for 3D point cloud analysis.
• The proposed model balances performance with significant reductions in computational and storage requirements.
• This work provides a lightweight, multi-branch graph network suitable for real-time 3D applications.