Action Recognition by Joint Spatial-Temporal Motion Feature

Table 3

Experiment on KTH dataset.

	Schindler and Van Gool CVPR’08 [20]; SNIPPET 1/SNIPPET 7	Jhuang et al. ICCV’07 [21]	Our approach
s1	90.9%/93.0%	96.0%	94.0%
s2	78.1%/81.1%	86.1%	88.3%
s3	88.5%/92.1%	89.8%	91.4%
s4	92.2%/96.7%	94.8%	93.8%