GOYA: Leveraging Generative Art for Content-Style Disentanglement

Area of Science:

• Computer Vision
• Artificial Intelligence
• Art History

Background:

• The content-style duality is key in art, with content referring to objects/concepts and style to visual appearance.
• Humans easily distinguish content and style, but computational methods struggle to capture this duality effectively.
• Existing methods like style transfer lack generalization, and classification is impractical due to style's spectrum.

Purpose of the Study:

• To develop a computational method for disentangling content and style in artworks.
• To leverage cutting-edge generative models for capturing artistic knowledge.
• To enable a more nuanced analysis of art by separating its constituent dimensions.

Main Methods:

• Introducing GOYA, a system designed to capture artistic knowledge from generative models.
• Utilizing generative models to explicitly learn and represent distinct artistic dimensions.
• Developing a novel approach for disentangling content and style in visual art.

Main Results:

• GOYA successfully learns to represent both content and style dimensions of artworks.
• The method demonstrates explicit disentanglement of artistic content and style.
• Experimental results validate GOYA's capability in capturing the content-style duality.

Conclusions:

• GOYA provides a robust framework for separating content and style in art.
• This work paves the way for utilizing generative models in art analysis.
• The findings offer new possibilities for computational art studies and understanding artistic representation.