Testing Deep Learning Models for Image Analysis Using Object-Relevant Metamorphic Relations

TL;DR

This paper introduces a metamorphic testing approach with novel relations to evaluate if deep learning models for image analysis rely on irrelevant features, revealing significant rates of inappropriate inferences.

Contribution

It proposes a new metamorphic testing method with two relations to detect irrelevant feature reliance in deep learning models for images.

Findings

Over 5.3% of image classification predictions are based on irrelevant features.

Object detection models show over 8.5% reliance on irrelevant features.

A new image generation strategy doubles attack success rates against models.

Abstract

Deep learning models are widely used for image analysis. While they offer high performance in terms of accuracy, people are concerned about if these models inappropriately make inferences using irrelevant features that are not encoded from the target object in a given image. To address the concern, we propose a metamorphic testing approach that assesses if a given inference is made based on irrelevant features. Specifically, we propose two novel metamorphic relations to detect such inappropriate inferences. We applied our approach to 10 image classification models and 10 object detection models, with three large datasets, i.e., ImageNet, COCO, and Pascal VOC. Over 5.3% of the top-5 correct predictions made by the image classification models are subject to inappropriate inferences using irrelevant features. The corresponding rate for the object detection models is over 8.5%. Based on the findings, we further designed a new image generation strategy that can effectively attack existing models. Comparing with a baseline approach, our strategy can double the success rate of attacks.