Fig. 1: The architecture for the EFTGAN model.

The green block is a ECNet model. The elemental convolution operation is used to extract the features of each element in the materials. The principal component analysis is used to downscale features to improve generator performance. The purple block is elemental features after downscaling. The blue block is InfoGAN model. The iteration use the multi-layer perceptron to predict the features generated by INfoGAN, returning the results to InfoGAN training.