Stochastic Object Models Designed for the Evaluation of Generative Adversarial Networks

Numerous realizations from three different stochastic object models (SOMs) are given as 256x256-pixel, 8-bit grayscale image files in PNG format. These images can be used as training data for a generative adversarial network (GAN). Each SOM was designed to exhibit specific contextual features that can be measured in a post-hoc analysis of images generated by the trained GAN. The goal is to quantify the rates at which a given GAN architecture fails to reproduce the prescribed features. More details about the SOMs and their use can be found in the manuscript pre-print: A Method for Evaluating the Capacity of Generative Adversarial Networks to Reproduce High-order Spatial Context SOM: Flags Eight different classes of "flags" are given. Each realization can be thought of as a grid of 16x16 pixel squares. Each class has precisely 80 squares set as bright foreground and 176 squares set as dark background. There are 24 squares that are never foreground in any class. The grayscale value of each pixel within the foreground is a random variate chosen from a scaled Beta distribution with parameters: a=4, b=2, location=96, scale=152. Analogously, each pixel within the background is a random variate chosen from a scaled Beta distribution with parameters: a=2, b=4, location=8, scale=192. All grayscale values throughout each realization are randomly arranged and were rounded to an integer value prior to placement. SOM: Voronoi Eight different classes of Voronoi tilings are given. Classes are distinguished by the total number of tiles (N) which ranges from 12 to 96, inclusively, in increments of 12. For each realization, N 8-bit grayscale shades are chosen from a uniform distribution of values running from 2 to 254, inclusively, in increments of 2. Each tile is given precisely one of the chosen shades. Crucially, the grayscale value of the tiles increases monotonically with tile area. In other words, the shade and area are perfectly rank-correlated in every realization. SOM: Alphabet An eight-letter alphabet comprises the letters H, K, L, V, W, X, Y, Z which where chosen for visual distinctiveness. A letter is equally sized and centered within a 32x32 pixel tile. The gray value of all letter pixels is 255 and the gray value of all background pixels is zero. Each 256x256 realization is made as a raster of 64 letters into an 8x8 grid where each letter is one of the 32x32-pixel tiles. For every realization, the frequencies of the letters is identical: 16-H, 2-K, 1-L, 4-V, 8-W, 1-X, 8-Y, and 24-Z. The order of letters is random wherever a higher-order letter-arrangement rule is not prescribed. For a pre-raster letter vector, these rules are: 1) V always and only occurs immediately after H 2) Y always and only occurs immediately after W 3) there must be at least 12 pairings of H next to Z, in either order. If any pairing happens to occur at the end of a grid row, the right-side partner is placed at the start of the next row below.