Continuous Gesture Sequences Recognition Based on Few-Shot Learning

Table 3

Model performance on Sebastien Marcel dynamic dataset.

Model	Similarity measure	Fine tuning	4-way 1-shot acc	4-way 3-shot acc
Full	MSE	N	89.62%	90.41%
	Cosine	N	91.77%	91.73%
Full	MSE	Y	91.14%	91.11%
	Cosine	Y	92.17%	93.23%