Fig. 1: GAN architecture. | npj Computational Materials

Fig. 1: GAN architecture.

From: Pores for thought: generative adversarial networks for stochastic reconstruction of 3D multi-phase electrode microstructures with periodic boundaries

Fig. 1: GAN architecture.

Schematic showing the architecture of the DC-GAN for 2D microstructural data. Generalisation to 3D samples is conceptually straightforward, but difficult to show as it requires the illustration of 4D tensors. In each layer, the green sub-volume shows a convolutional kernel at an arbitrary location in the volume and the blue sub-volume is the result of that convolution. In each case, the kernel is the same depth as the one non-spatial dimension, c, but must scan through the two spatial dimensions in order to build up the image in the following layer.

Back to article page