Table 3.

Recognition results of different classification models.

Model	Backbone	Accuracy (%)	F1 Score (%)	Params (M)	Speed (ms)
C3D + Linear Classifier	C3D	88.92	88.82	78.41	1264
CNN + LSTM	ResNet50	91.52	91.49	24.30	29
TSN	ResNet50	91.62	91.56	23.52	635
R2plus1D	3D ResNet34	92.13	92.16	63.76	1104
I3D	3D ResNet50	93.32	93.29	28.04	116
ResNeXt50-T	ResNeXt50	95.69	95.65	22.98	29