Table 1 Comparison of our model performance using Sigmoid and Tanh
From: CNN-optimized text recognition with binary embeddings for Arabic expiry date recognition
Input size | No. of samples train/test | Inference Speed (img/s) | Convergence @epoch | Activation function | Test accuracy (%) |
|---|---|---|---|---|---|
256 × 64 × 3 | 2629/658 | 1 k | 50 | Sigmoid | 95.59 |
256 × 64 × 3 | 2629/658 | 1 k | 50 | Tanh | 98.94 |