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

Divyansh Jha, Kai Yi, Ivan Skorokhodov, Mohamed Elhoseiny

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

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/

Original languageEnglish (US)
Title of host publicationProceedings of the 13th International Conference on Computational Creativity, ICCC 2022
EditorsMaria M. Hedblom, Anna Aurora Kantosalo, Roberto Confalonieri, Oliver Kutz, Tony Veale
PublisherAssociation for Computational Creativity (ACC)
Pages195-204
Number of pages10
ISBN (Electronic)9789895416042
StatePublished - 2022
Event13th International Conference on Computational Creativity, ICCC 2022 - Bozen-Bolzano, Italy
Duration: Jun 27 2022Jul 1 2022

Publication series

NameProceedings of the 13th International Conference on Computational Creativity, ICCC 2022

Conference

Conference13th International Conference on Computational Creativity, ICCC 2022
Country/TerritoryItaly
CityBozen-Bolzano
Period06/27/2207/1/22

ASJC Scopus subject areas

  • Computational Theory and Mathematics

Fingerprint

Dive into the research topics of 'Creative Walk Adversarial Networks: Novel Art Generation with Probabilistic Random Walk Deviation from Style Norms'. Together they form a unique fingerprint.

Cite this