Verifiability and Predictability: Interpreting Utilities of Network Architectures for Point Cloud Processing

TL;DR

This paper investigates how different intermediate-layer architectures in deep neural networks affect their ability to process 3D point clouds, proposing metrics and revisions to enhance their utility and robustness.

Contribution

It introduces five novel metrics for diagnosing DNNs' intermediate layers and uses these to improve network architectures for better point cloud processing.

Findings

Metrics effectively diagnose network utilities

Revised architectures show improved robustness

Hypotheses are validated through experiments

Abstract

In this paper, we diagnose deep neural networks for 3D point cloud processing to explore utilities of different intermediate-layer network architectures. We propose a number of hypotheses on the effects of specific intermediate-layer network architectures on the representation capacity of DNNs. In order to prove the hypotheses, we design five metrics to diagnose various types of DNNs from the following perspectives, information discarding, information concentration, rotation robustness, adversarial robustness, and neighborhood inconsistency. We conduct comparative studies based on such metrics to verify the hypotheses. We further use the verified hypotheses to revise intermediate-layer architectures of existing DNNs and improve their utilities. Experiments demonstrate the effectiveness of our method.