Visual Detection of Generic Defects in Industrial Components using Generative Adversarial Networks

Detection of general or unspecified anomalies in industrial environments using camera images is a challenging task, and one which is the focus of much current research. Many studies are application focused, and demonstrate results on domain-specific datasets which are often not widely available, and hence it is difficult to directly compare performance of new techniques. We propose the use of Generative Adversarial Networks (GAN)s in performing generalised visual anomaly detection and the performance of various state-of-the-art GAN based methods is evaluated and compared to previous results. A simple method for generation of a dataset with automatically annotated anomalies in the form of Gaussian blobs in images, demonstrated on the well-known MNIST hand-written digits dataset, in order that unsupervised visual anomaly detection techniques can be compared without access to industry-specific datasets.