README.md

metadata

title: MNIST Digit Recognition
emoji: 🎯
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.3.0
app_file: app.py
pinned: false
license: mit

🎯 MNIST Digit Recognition Demo

Draw a digit (0-9) and watch the AI recognize it with 99.6% accuracy!

About This Model

This Space demonstrates a Convolutional Neural Network trained on the MNIST dataset, achieving exceptional performance:

• Test Accuracy: 99.60%
• Model Size: 271K parameters
• Architecture: 4-layer CNN with batch normalization
• Framework: PyTorch

How to Use

1. Draw a digit in the canvas on the left
2. Wait a moment for the prediction
3. See the result with confidence scores for all digits

The model was trained using advanced techniques including:

• Data augmentation (rotation, scaling, random erasing)
• OneCycleLR scheduler with warmup
• Dropout and batch normalization
• Early stopping

Model Repository

Full training code and model weights: mnist-cnn-classifier

Performance

The model achieves near-perfect accuracy across all digits:

• Most digits: 99-100% accuracy
• Balanced performance (no digit is significantly harder)
• Fast inference (~5ms per image on CPU)

Try it out and see if you can draw digits the model gets wrong! 🎨