A Chinese Lip-Reading System Based on Convolutional Block Attention Module

Table 1

The accuracy of each model.

Model	Prams (M)	Top-1 accuracy (%)
Vgg16 + LSTM + Attention	144.4	95.2
Vgg16 + GRU + Attention	142.4	95.3
InceptionV3 + LSTM + Attention	31.9	98.2
InceptionV3 + GRU + Attention	29.9	99.1
ResNet50 + LSTM + Attention	33.6	98.2
ResNet50 + GRU + Attention	31.6	99.3
ResNet101 + LSTM + Attention	52.6	97.3
ResNet101 + GRU + Attention	50.6	99.6
ResNet152 + LSTM + Attention	68.3	98.4
ResNet152 + GRU + Attention	66.3	99.8
ResNet50 + CBAM + LSTM + Attention	36.1	98.7
ResNet50 + CBAM + GRU + Attention	34.1	99.6