If you got here from the QR code on the poster I made for sign-lang@LREC 2026, thanks for looking at my poster!
Click here to download the paper, or you can also email me at fredchan@uw.edu.
A Small Model for Big Articulators: Sign Language Detection With a Tiny Machine Learning Model describes a 1,013 convolutional neural network used for clipping big videos containing sequences of isolated signs into individual clips containing one sign each.
Its intended use case is compiling dictionaries and making stimuli for psycho-/neuro-linguistics studies. Instead of manually clipping a recording into hundreds or even thousands of individual sign videos, our model can save time by detecting frames containing signing while using a tiny amount of compute.
