CNN_best_model

Pre-trained CNN Weights (MNIST)

This release contains the best-performing model weights (CNN_best_model.pth) for the Handwritten Digit Recognition System. These weights allow users to run inference or evaluation immediately without needing to retrain the model from scratch.

---

Performance Metrics

The model was trained on the MNIST training set (with data augmentation) and evaluated on the official test set (10,000 images).

Metric	Value
Test Accuracy	99.70%
Test Loss	0.0884
Parameters	~3.36M

---

Usage Instructions

1. Download CNN_best_model.pth from the Assets section below.
2. Place the file in the root directory of the project (Digit-Recognition/).
3. Run the notebook or script. The training pipeline will automatically detect the file and skip the training phase:

   # Expected output:
   Found existing model: 'CNN_best_model.pth'
   Loading weights and skipping training...

---

Generated by the main.ipynb pipeline on Apple Silicon (M3 Pro)