Continuous Gesture Sequences Recognition Based on Few-Shot Learning

Table 1

Model performance on RWTH dataset.

Model	Similarity measure	Fine tuning	5-way 1-shot acc	5-way 3-shot acc
Full	MSE	N	77.40%	78.39%
Full	Cosine	N	86.67%	88.20%
Full	MSE	Y	79.45%	80.72%
Full	Cosine	Y	89.73%	90.16%