Stanford convolutional neural networks for visual recognition