SwiDeN : Convolutional Neural Networks For Depiction Invariant Object Recognition

Current state of the art object recognition architectures achieve impressiveperformance but are typically specialized for a single depictive style (e.g.photos only, sketches only). In this paper, we present SwiDeN : ourConvolutional Neural Network (CNN) architecture which recognizes objectsregardless of how they are visually depicted (line drawing, realistic shadeddrawing, photograph etc.). In SwiDeN, we utilize a novel `deep' depictivestyle-based switching mechanism which appropriately addresses thedepiction-specific and depiction-invariant aspects of the problem. We compareSwiDeN with alternative architectures and prior work on a 50-category Photo-Artdataset containing objects depicted in multiple styles. Experimental resultsshow that SwiDeN outperforms other approaches for the depiction-invariantobject recognition problem.