Conditional GANs For Painting Generation
This work addresses the problem of generating customizable art for creative applications, but it is incremental as it builds on existing GAN methods with conditioning.
The study tackled generating novel oil paintings using Generative Adversarial Networks, finding that Spectral Normalization GAN produced the most comparable results to the training dataset based on visual and quantitative metrics like Sliced Wasserstein Distance, and developed a novel architecture for generating face paintings with user-specified characteristics.
We examined the use of modern Generative Adversarial Nets to generate novel images of oil paintings using the Painter By Numbers dataset. We implemented Spectral Normalization GAN (SN-GAN) and Spectral Normalization GAN with Gradient Penalty, and compared their outputs to a Deep Convolutional GAN. Visually, and quantitatively according to the Sliced Wasserstein Distance metric, we determined that the SN-GAN produced paintings that were most comparable to our training dataset. We then performed a series of experiments to add supervised conditioning to SN-GAN, the culmination of which is what we believe to be a novel architecture that can generate face paintings with user-specified characteristics.