MNIST-like datasets classifier

Convolutional neural network to recognize symbols using a MNIST-like datasets

Purpose

Recognize handwritten letters and digits

Accuracy

92%

Requirements

PyTorch, TorchVision, Numpy (1.23.1), matplotlib.pyplot

Requirements for datasets

28x28 grayscaled images

Tested datasets

• MNIST
• EMNIST

Run

python3 cnn.py

Runtime options

# Comment or uncomment necessary functions
train(model, model_path)
test(model_path)
test_all(model_path)

Warning

Training NN or testing all images (function test_all()) may take very long time.

Output example