Fig. 1From: Automatic differentiation of Glaucoma visual field from non-glaucoma visual filed using deep convolutional neural networkDiagram showing the modified VGG network. VGG15 was adopted as our network structure. We modified the output dimension of the penultimate layer fc7 from 4096 to 200. And the last layer is modified to output a two-dimension vector which corresponds to the prediction scores of healthy VF and glaucoma VF. The network is first pre-trained on a large scale, natural image classification dataset ImageNet16 to initialize its parameters. Then we modified the last two layers as mentioned above and initialized their parameters by drawing from a Gaussian distributionBack to article page