Gestures

Recurrent neural networks (RNN) require large training datasets from which they learn new class models. This limitation prohibits their …

Some gestural input is accidental or malformed. For such cases, a score similarity threshold must be set, and all input that scores below that threshold will be rejected. We provide a technique for automatically selecting such a threshold for custom gestures.

In a continuous stream of data, the starting and ending points of gestures are not known. We present Machete - an approach to identifying such points in the data, solving the segmentation problem.

When playing video games, users input gestures that are more variable than those collected in lab conditions. Systems trained on lab data will underperform in real-world scenarios. We designed a simple-to-implement protocol which helps elicit more variability during user gesture collection.