Variance-Aware Weight Initialization for Point Convolutional Neural Networks

TL;DR

This paper introduces a variance-aware weight initialization method for point convolutional neural networks that eliminates the need for batch normalization, especially beneficial for small batch sizes in point cloud processing.

Contribution

It proposes a unified framework for continuous convolutions and a novel weight initialization strategy tailored for point CNNs, addressing the limitations of batch normalization.

Findings

Variance-aware initialization matches or outperforms batch normalization.

The method is effective for small batch sizes in point cloud learning.

Unifies various continuous convolution approaches.

Abstract

Appropriate weight initialization has been of key importance to successfully train neural networks. Recently, batch normalization has diminished the role of weight initialization by simply normalizing each layer based on batch statistics. Unfortunately, batch normalization has several drawbacks when applied to small batch sizes, as they are required to cope with memory limitations when learning on point clouds. While well-founded weight initialization strategies can render batch normalization unnecessary and thus avoid these drawbacks, no such approaches have been proposed for point convolutional networks. To fill this gap, we propose a framework to unify the multitude of continuous convolutions. This enables our main contribution, variance-aware weight initialization. We show that this initialization can avoid batch normalization while achieving similar and, in some cases, better performance.