A Study and Comparison of Human and Deep Learning Recognition Performance Under Visual Distortions
This highlights a robustness gap in DNNs for computer vision, which could guide development of more robust models.
The study compared deep neural networks (DNNs) and human subjects on image classification under distortions like blur and noise, finding that DNNs perform much worse than humans on distorted images and show little correlation in errors.
Deep neural networks (DNNs) achieve excellent performance on standard classification tasks. However, under image quality distortions such as blur and noise, classification accuracy becomes poor. In this work, we compare the performance of DNNs with human subjects on distorted images. We show that, although DNNs perform better than or on par with humans on good quality images, DNN performance is still much lower than human performance on distorted images. We additionally find that there is little correlation in errors between DNNs and human subjects. This could be an indication that the internal representation of images are different between DNNs and the human visual system. These comparisons with human performance could be used to guide future development of more robust DNNs.