Digit Recognition System

Features

• MNIST-based CNN model
• Real-time digit segmentation
• Confidence scoring
• Image preprocessing pipeline

Complete Workflow Guide

1. System Setup

# Clone repository
git clone https://github.com/TrendoD/digit-recognition
cd digit-recognition

# Set up environment
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
mkdir data\raw models\saved_models results

2. Model Training

# Run training with default parameters
python src/model/train_model.py

# Optional arguments for advanced users:
# python src/model/train_model.py --epochs 20 --batch_size 64

3. Model Evaluation

# Run comprehensive evaluation
python src/model/evaluate_model.py

# Results will be saved in:
# - results/evaluation/confusion_matrix.png
# - Console output shows precision/recall metrics

4. Making Predictions

Only 1 Image Prediction:

Run with:

python digit_recognizer.py --image_path path/to/your_image.png

Direktori Full Predictions :

Run with :

python digit_recognizer.py --dir path/to/your_folder

Troubleshooting

Q: Getting "No such file" errors? A: Ensure:

1. Model file exists in models/saved_models/
2. Test images are in data/raw/
3. All directories are created

Q: Low confidence predictions? A: Ensure input images:

• Have clear contrast
• Digits are centered
• Background is uniform
• Image size > 100x50 pixels