3D Point Cloud Feature Explanations Using Gradient-Based Methods

TL;DR

This paper extends gradient-based explainability methods to 3D point cloud data, analyzing feature importance and demonstrating significant model pruning with minimal accuracy loss.

Contribution

It introduces a model-agnostic approach for explaining 3D neural network features and shows how sparsity can be exploited for efficient pruning.

Findings

Edges and corners are identified as important features in 3D data.

Voxel-based networks can be pruned to 5% of parameters with negligible accuracy loss.

Features in 3D data are inherently sparse and amenable to pruning.

Abstract

Explainability is an important factor to drive user trust in the use of neural networks for tasks with material impact. However, most of the work done in this area focuses on image analysis and does not take into account 3D data. We extend the saliency methods that have been shown to work on image data to deal with 3D data. We analyse the features in point clouds and voxel spaces and show that edges and corners in 3D data are deemed as important features while planar surfaces are deemed less important. The approach is model-agnostic and can provide useful information about learnt features. Driven by the insight that 3D data is inherently sparse, we visualise the features learnt by a voxel-based classification network and show that these features are also sparse and can be pruned relatively easily, leading to more efficient neural networks. Our results show that the Voxception-ResNet model can be pruned down to 5\% of its parameters with negligible loss in accuracy.