A Deep Convolutional Neural Networks based classifier to classify old Japanese script.
Kuzushiji-MNIST is a drop-in replacement for the MNIST dataset (28x28 grayscale, 70,000 images), provided in the original MNIST format as well as a NumPy format. Since MNIST restricts us to 10 classes, we chose one character to represent each of the 10 rows of Hiragana when creating Kuzushiji-MNIST.
By non resnet architectures, we were able to achieve the accuracy of 97.18%.
Layer (type) Output Shape Parameters
input_1 (InputLayer) (None, 28, 28, 1) 0
zero_padding2d_1 (ZeroPaddin (None, 32, 32, 1) 0
conv2d_1 (Conv2D) (None, 26, 26, 32) 1568
batch_normalization_1 (Batch (None, 26, 26, 32) 128
max_pooling2d_1 (MaxPooling2 (None, 13, 13, 32) 0
activation_1 (Activation) (None, 13, 13, 32) 0
conv2d_2 (Conv2D) (None, 9, 9, 64) 51200
batch_normalization_2 (Batch (None, 9, 9, 64) 256
activation_2 (Activation) (None, 9, 9, 64) 0
conv2d_3 (Conv2D) (None, 4, 4, 128) 73728
batch_normalization_3 (Batch (None, 4, 4, 128) 512
max_pooling2d_2 (MaxPooling2 (None, 2, 2, 128) 0
activation_3 (Activation) (None, 2, 2, 128) 0
flatten_1 (Flatten) (None, 512) 0
dense_1 (Dense) (None, 512) 262656
dropout_1 (Dropout) (None, 512) 0
dense_2 (Dense) (None, 10) 5130
Total params: 395,178 Trainable params: 394,730 Non-trainable params: 448