Table 2.

Accuracy of different deep learning models as deep feature extractor in 9-class and 4-class sleep posture classification, and the number of parameters of each model.

Model	Nine-Class Fine-Grained	Four-Class Coarse-Grained	Number of Parameters
ResNet-50	0.496	0.721	2.05B
EfficientNet-B0	0.454	0.775	42.55M
DenseNet-121	0.534	0.804	64.02M
PHResNet-50	0.468	0.723	2.08B
Attention-56	NC	0.469	4.62B
Swin Transformer	NC	NC	11.06B

NC: model did not converge.