SensoryArt

The automatic recognition of sensory gestures in artworks provides the opportunity to open up methods of computational humanities to modern paradigms like sensory studies or everyday history.

We introduce SensoryArt, a dataset of multisensory gestures in historical artworks, annotated with person boxes, pose estimation key points and gesture labels.

We analyze algorithms for each label type and explore their combination for gesture recognition without intermediate supervision. These combined algorithms are evaluated for their ability to recognize and localize depicted persons performing sensory gestures. Our experiments show that direct detection of smell gestures is the most effective method for both detecting and localizing gestures. After applying post-processing, this method outperforms even image-level classification algorithms in image-level classification metrics, despite not being the primary training objective.

This work aims to open up the field of sensory history to the computational humanities and provide humanities-based scholars with a solid foundation to complement their methodological toolbox with quantitative methods.

Instruction on how to use the dataset for training and evaluating recognition algorithms can be found on GitHub: https://github.com/mathiaszinnen/SensoryGestureRecognition