MNIST Drawing Recognizer

A lightweight CNN model for real-time handwritten digit recognition (0-9) with a PyQt5 drawing interface (see full project).

Model Description

• Architecture: Simple CNN with 2 convolutional layers + fully connected layers
• Framework: PyTorch
• Output: Digit 0-9 + confidence score
• Accuracy: ~98.7% on MNIST test set

Source Code

Full project: Click here!

Usage

from huggingface_hub import hf_hub_download
import torch
from train import Net

model_path = hf_hub_download("monadayzek/mnist-drawing-recognizer", "mnist_cnn.pth")

model = Net().to('cpu')
model.load_state_dict(torch.load(model_path, map_location='cpu', weights_only=True))
model.eval()