The involvement of AI in our day to day uses has increased over time. One of those involvements includes art. AI has been helping artists generate songs, logos and many more. It looks like a developer has managed to use the power of AI in the manipulation of images.
CycleGAN Adjust The Images
Developer Holy Grimm, Open AI Scholar program graduate created the AI model using Generative Adversarial Network (GAN). The GAN consists of a data generator and a discriminator. Grimm chose CycleGAN which learns transformations among two images.
CycleGAN takes a set of images and converts them to another set of visually similar images. The CycleGAN uses image-to-image comparison to compare the images and try to make it another kind of image. The GAN has been already tested to make horses look like zebras and apple look like an orange.
Grimm used ResNet50 algorithm to further improvise her AI. As result, we got Art Composition Attributes Network (ACAN). ACAN learned to started producing images by varying eight compositions of them image namely – shape, size, color, texture, primary color, contrast, repetition, and color harmony.
Grimm said she distinguished eight composition attributes of the image using her ACAN only by using a sample size of 500 images. She added that her AI can use the heat map to distinguish and learn association with between colors and the color wheel.
Grimm graduated from Open AI Scholars program which happens to help out people who are underrated in their fields. This organization based in San Francisco is backed by Elon Musk and many tech dignitaries.