Artistic Domain Generalisation Methods are Limited by their Deep Representations
This work addresses the problem of domain generalization in computer vision for researchers, revealing limitations in current methods and suggesting a need for higher-order representations like structure and shape.
The paper tackles the cross-depiction problem in visual object recognition by showing that fixing the last layer of AlexNet to random values achieves performance comparable to state-of-the-art domain adaptation and generalization methods on the PACS benchmark, indicating that texture alone is insufficient for generalization.
The cross-depiction problem refers to the task of recognising visual objects regardless of their depictions; whether photographed, painted, sketched, {\em etc}. In the past, some researchers considered cross-depiction to be domain adaptation (DA). More recent work considers cross-depiction as domain generalisation (DG), in which algorithms extend recognition from one set of domains (such as photographs and coloured artwork) to another (such as sketches). We show that fixing the last layer of AlexNet to random values provides a performance comparable to state of the art DA and DG algorithms, when tested over the PACS benchmark. With support from background literature, our results lead us to conclude that texture alone is insufficient to support generalisation; rather, higher-order representations such as structure and shape are necessary.