DeepCorrect: Correcting DNN models against Image Distortions
share
In recent years, the widespread use of deep neural networks (DNNs) has facilitated great improvements in performance for computer vision tasks like image classification and object recognition. In most realistic computer vision applications, an input image undergoes some form of image distortion such as blur and additive noise during image acquisition or transmission. Deep networks trained on pristine images perform poorly when tested on distorted images affected by image blur or additive noise. In this paper, we evaluate the effect of image distortions like Gaussian blur and additive noise on the outputs of pre-trained convolutional filters. We propose a metric to identify the most noise susceptible convolutional filters and rank them in order of the highest gain in classification accuracy upon correction. In our proposed approach called DeepCorrect, we apply small convolutional filter blocks on top of these ranked filters and train them to correct the worst noise and blur affected filter outputs. Applying DeepCorrect on the CIFAR-100 dataset, we significantly improve the robustness of DNNs against distorted images and also outperform the alternative approach of fine-tuning networks.
READ FULL TEXT