Explorations in Texture Learning

Blaine Hoak, Patrick McDaniel

Keywords: texture bias, explainability, interpretability, learned patterns

TL;DR: We study how object classification models respond to specific textures, and how this information uncovers properties about what models have learned.

Abstract: In this work, we investigate *texture learning*: the identification of textures learned by object classification models, and the extent to which they rely on these textures. We build texture-object associations that uncover new insights about the relationships between texture and object classes in CNNs and find three classes of results: associations that are strong and expected, strong and not expected, and expected but not present. Our analysis demonstrates that investigations in texture learning enable new methods for interpretability and have the potential to uncover unexpected biases.

Submission Number: 56