Testing Deep Learning Models for Image Analysis Using Object-Relevant Metamorphic Relations
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.
Cite
@article{arxiv.1909.03824,
title = {Testing Deep Learning Models for Image Analysis Using Object-Relevant Metamorphic Relations},
author = {Yongqiang Tian and Shiqing Ma and Ming Wen and Yepang Liu and Shing-Chi Cheung and Xiangyu Zhang},
journal= {arXiv preprint arXiv:1909.03824},
year = {2021}
}
Comments
Please note that a later version of this paper is accepted by Empirical Software Engineering in 2021. The title of the accepted paper is: "To What Extent Do DNN-based Image Classification Models Make Unreliable Inferences?". Please contact the first author if you are interested in the accepted version