3D Point Cloud Network Pruning: When Some Weights Do not Matter

Amrijit Biswas,Md. Ismail Hossain,M M Lutfe Elahi,Ali Cheraghian,Fuad Rahman,Nabeel Mohammed,Shafin Rahman

2024-08-27

Abstract:A point cloud is a crucial geometric data structure utilized in numerous applications. The adoption of deep neural networks referred to as Point Cloud Neural Networks (PC- NNs), for processing 3D point clouds, has significantly advanced fields that rely on 3D geometric data to enhance the efficiency of tasks. Expanding the size of both neural network models and 3D point clouds introduces significant challenges in minimizing computational and memory requirements. This is essential for meeting the demanding requirements of real-world applications, which prioritize minimal energy consumption and low latency. Therefore, investigating redundancy in PCNNs is crucial yet challenging due to their sensitivity to parameters. Additionally, traditional pruning methods face difficulties as these networks rely heavily on weights and points. Nonetheless, our research reveals a promising phenomenon that could refine standard PCNN pruning techniques. Our findings suggest that preserving only the top p% of the highest magnitude weights is crucial for accuracy preservation. For example, pruning 99% of the weights from the PointNet model still results in accuracy close to the base level. Specifically, in the ModelNet40 dataset, where the base accuracy with the PointNet model was 87. 5%, preserving only 1% of the weights still achieves an accuracy of 86.8%. Codes are available in: <a class="link-external link-https" href="https://github.com/apurba-nsu-rnd-lab/PCNN_Pruning" rel="external noopener nofollow">this https URL</a>

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to solve the problem of excessive computational and memory requirements faced by 3D point - cloud neural networks (PCNNs) when processing large - scale 3D point - cloud data. Specifically, with the expansion of neural network models and the scale of 3D point - cloud data, how to effectively reduce the consumption of computational resources, reduce latency and maintain model performance has become an important research topic.

#### Main problems:
1. **Excessive computational and memory requirements**: Expanding the scale of neural network models and 3D point - cloud data will significantly increase computational and memory requirements, which poses a challenge to the low - energy - consumption and low - latency requirements in practical applications.
2. **Redundant parameters**: Traditional PCNNs are very sensitive to parameters, making it difficult to effectively reduce redundant weights through traditional pruning methods.
3. **Effectiveness of pruning techniques**: Existing pruning techniques have not been fully explored in the 3D field. Especially for high - dimensional and geometrically complex data, how to find effective pruning strategies is an urgent problem to be solved.

#### Research objectives:
- **Verify the "Lottery Hypothesis" (LTH)**: Explore whether there are sparse sub - networks (winning tickets) that can maintain or exceed the performance of the original model while significantly reducing the number of parameters.
- **Develop efficient pruning methods**: Propose pruning strategies applicable to 3D point - cloud neural networks, such as Iterative Magnitude Pruning (IMP) and one - shot global pruning, to achieve efficient compression.
- **Evaluate the effect of transfer learning**: Study whether the sparse sub - networks obtained from a 3D shape classification task can be transferred to other related tasks and maintain high accuracy.

#### Specific problems:
1. **Can winning tickets be found?** In an over - parameterized 3D shape classification network, can sparse sub - networks that maintain or exceed the performance of the original network be found?
2. **Optimal pruning strategy**: What pruning strategies and techniques are most effective in view of the unique challenges of 3D models?
3. **Transferability**: Can the winning tickets obtained from a certain 3D shape classification task or dataset be transferred to different but related tasks while maintaining high accuracy?

Through the research of these problems, the authors hope to provide new insights and solutions for the field of 3D model compression, especially to support efficient inference on edge devices.