elangbp/cnn-digits-project

CNN handwritten digit classifier

A PyTorch CNN trained on the scikit-learn digits dataset. The 8x8 grayscale digits are converted to RGB for model input. Checkpoint file: best_digits_cnn.pth

• Test accuracy: 93.33%
• Labels: digits 0 through 9
• Input normalization: mean (0.5, 0.5, 0.5), standard deviation (0.5, 0.5, 0.5) Recreate the CNN architecture from the Colab notebook before loading the checkpoint with load_state_dict.