comparison of small sample methods for Handshape Recognition

Automatic Sign Language Translation (SLT) systems can be a great asset to improve the communication with and within deaf communities. Currently, main issue preventing effective translation models lays in low availability of labelled data, which hinders use modern deep learning models.
 SLT is complex problem that involves many subtasks, handshape recognition most important. We compare series specially tailored for small datasets their performance on tasks. evaluate Wide-DenseNet few-shot Prototypical Network without transfer learning, also using Model-Agnostic Meta-Learning (MAML).
 Our findings indicate Prototipical Networks provide best results. networks, particularly, are vastly superior when less than 30 samples, while achieves results more samples. On other hand, MAML does not any scenario. These help design better models.