Table 6 Performance of Computational Costs for the Proposed AI Models.

From: Integrating CNN and transformer architectures for superior Arabic printed and handwriting characters classification

The AI models

Fine-tune layers

Trainable params

Training time (msec)

Average training time/ epoch

VGG16

17

2,889,218

1050.61

42.60

ResNet50

123

50,557,404

1445.54

57.83

Ensemble model

Concatenated layers

53,446,622

2294.23

91.76

The proposed hybrid transformer model

1-vector transformation

92,438,978

2963.34

118.53

Back to article page