Figure 13.

Four batches of 100 sample generation results trained with a 2-D mixture of Gaussian by: (a) BEGAN; (b) SFGAN with $\alpha =\mathrm{12,288}$; (c) training data.