In this work we investigate the effect of the convolutional network depth onits accuracy in the large-scale image recognition setting. Our maincontribution is a thorough evaluation of networks of increasing depth using anarchitecture with very small (3x3) convolution filters, which shows that asignificant improvement on the prior-art configurations can be achieved bypushing the depth to 16-19 weight layers. These findings were the basis of ourImageNet Challenge 2014 submission, where our team secured the first and thesecond places in the localisation and classification tracks respectively. Wealso show that our representations generalise well to other datasets, wherethey achieve state-of-the-art results. We have made our two best-performingConvNet models publicly available to facilitate further research on the use ofdeep visual representations in computer vision.