ACAN neural network is able to adjust the contrast, size and shape of images

A graduate of the Open AI Scholar program, Holi Grim, has created her own neural network model that can modify images. This modification creates new images from the source.

Grim called her project Art Composition Attributes Network (ACAN). The neural network changes shape, size, color and its combinations, texture, background, contrasts and repetitions. Grim revealed these eight parameters when she worked with a sample of 500 images, testing her design. ACAN also works with color maps, examines the relationship and differences in colors and shades of the entire spectrum.

At the core of Grim’s work is the Generative-Adversary Network (GAN). It changes images to visually similar, making them additions. For example, a horse can turn it into a zebra, and an apple into an orange.

robo-hunter.com