Systematic Testing of Convolutional Neural Networks for Autonomous Driving

TL;DR

This paper introduces a systematic framework for analyzing CNNs in autonomous driving, using synthetic image generation and visualization tools to test and understand model vulnerabilities and performance.

Contribution

The paper presents a novel framework combining synthetic image generation and visualization for comprehensive CNN analysis in autonomous vehicle applications.

Findings

Identifies CNN vulnerabilities through synthetic testing.

Enables comparison of different classification models.

Facilitates dataset generation for training and validation.

Abstract

We present a framework to systematically analyze convolutional neural networks (CNNs) used in classification of cars in autonomous vehicles. Our analysis procedure comprises an image generator that produces synthetic pictures by sampling in a lower dimension image modification subspace and a suite of visualization tools. The image generator produces images which can be used to test the CNN and hence expose its vulnerabilities. The presented framework can be used to extract insights of the CNN classifier, compare across classification models, or generate training and validation datasets.