A Methodology to Identify Cognition Gaps in Visual Recognition Applications Based on Convolutional Neural Networks

TL;DR

This paper introduces a methodology to identify cognition gaps in CNN-based visual recognition systems by generating worst-case images through adversarial search, highlighting potential weaknesses in their understanding.

Contribution

It presents a novel approach combining image augmentation and adversarial search to detect cognitive gaps in CNNs, demonstrated on AlexNet for driving scenarios.

Findings

Identified cognition gaps in CNNs using worst-case images

Effective adversarial search generates challenging images

Method applicable to safety-critical applications

Abstract

Developing consistently well performing visual recognition applications based on convolutional neural networks, e.g. for autonomous driving, is very challenging. One of the obstacles during the development is the opaqueness of their cognitive behaviour. A considerable amount of literature has been published which describes irrational behaviour of trained CNNs showcasing gaps in their cognition. In this paper, a methodology is presented that creates worstcase images using image augmentation techniques. If the CNN's cognitive performance on such images is weak while the augmentation techniques are supposedly harmless, a potential gap in the cognition has been found. The presented worst-case image generator is using adversarial search approaches to efficiently identify the most challenging image. This is evaluated with the well-known AlexNet CNN using images depicting a typical driving scenario.