Creative Walk Adversarial Networks: Novel Art Generation with Probabilistic Random Walk Deviation from Style Norms

We propose Creative Walk Adversarial Networks (CWAN) for novel art generation. Quality learning representation of unseen art styles is critical to facilitate generation of new meaningful artworks. CWAN learns an improved metric space for generative art by exploring unseen visual spaces with probabilistic random walks. CWAN constructs a dynamic graph that includes the seen art style centers and generated samples in the current minibatch. We then initiate a random walk from each art style center through the generated artworks in the current minibatch. As a deviation signal, we encourage the random walk to eventually land after T steps in a feature representation that is difficult to classify as any of the seen art styles. We investigate the ability of the proposed loss to generate meaningful novel visual art on the WikiArt dataset. Our experimental results and human evaluations demonstrate that CWAN can generate novel art that is significantly more preferable compared to strong state-of-the-art methods, including StyleGAN2 and StyleCAN2. The code is publicly available at: https://vision-cair.github.io/CWAN/