LeNet-5 MNIST & EMNIST Balanced Classifier A LeNet-5 CNN trained on the MNIST and EMNIST Balanced datasets for handwritten character recognition.
Training Details for MNIST Dataset
Metric
Score
Weighted F1
0.9914
Val Accuracy
~0.99
Classes
10
Dataset
MNIST
Training Details for EMNIST Dataset
Metric
Score
Weighted F1
0.8751
Val Accuracy
~0.88
Classes
47
Dataset
EMNIST Balanced
EMNIST Balanced Training Curves
Layer
Output Shape
Params
Conv2D (6 filters, 5×5, ReLU, same)
28×28×6
156
AveragePooling2D (2×2)
14×14×6
0
Conv2D (16 filters, 5×5, ReLU)
10×10×16
2,416
AveragePooling2D (2×2)
5×5×16
0
Flatten
400
0
Dropout (0.2)
400
0
Dense (120, ReLU)
120
48,120
Dense (84, ReLU)
84
10,164
Dense (10, Softmax)
10
850
Total 61,706
LeNet-5 EMNIST Balanced (modified)
Layer
Output Shape
Params
Conv2D (6 filters, 5×5, ReLU, same)
28×28×6
156
AveragePooling2D (2×2)
14×14×6
0
Conv2D (16 filters, 5×5, ReLU)
10×10×16
2,416
AveragePooling2D (2×2)
5×5×16
0
Flatten
400
0
Dropout (0.2)
400
0
Dense (120, ReLU)
120
48,120
Dropout (0.2)
120
0
Dense (84, ReLU)
84
10,164
Dense (47, Softmax)
47
3,995
Total 64,851