Spaces:
Sleeping
Sleeping
|
Download README.md from EnYa32/DigitRecognitionCNN: direct link, hf CLI and curl.
- Browser
- Download file 1.48 kB
-
https://huggingface.co/spaces/EnYa32/DigitRecognitionCNN/resolve/main/README.md
- Command line
-
hf download hf://spaces/EnYa32/DigitRecognitionCNN/README.md
-
curl -L -o README.md https://huggingface.co/spaces/EnYa32/DigitRecognitionCNN/resolve/main/README.md
1.48 kB
| title: DigitRecognitionCNN | |
| emoji: ✍️ | |
| colorFrom: red | |
| colorTo: red | |
| sdk: streamlit | |
| app_file: src/streamlit_app.py | |
| pinned: false | |
| short_description: Handwritten digit recognition (0–9) using a CNN and MNIST. | |
| license: mit | |
| # ✍️ MNIST Digit Recognizer (CNN) | |
| This project is a simple Computer Vision application that recognizes handwritten digits (0–9) using a Convolutional Neural Network (CNN). | |
| ## Dataset | |
| - Kaggle Competition: **Digit Recognizer (MNIST)** | |
| - Each image is **28×28 grayscale** (784 pixels). | |
| - Labels are digits **0–9**. | |
| ## Approach | |
| 1. Load `train.csv` and `test.csv` | |
| 2. Normalize pixel values to **[0, 1]** | |
| 3. Reshape to **(28, 28, 1)** | |
| 4. Train a CNN with early stopping | |
| 5. Evaluate using validation accuracy and confusion matrix | |
| 6. Generate `submission.csv` for Kaggle | |
| ## Results | |
| - The model achieved around **~99% validation accuracy**. | |
| - Most digits are predicted correctly; errors mostly happen for visually similar digits. | |
| ## Streamlit App | |
| The app lets you upload an image of a handwritten digit and outputs: | |
| - predicted digit | |
| - confidence score | |
| - probability distribution for all 10 digits | |
| - preprocessed 28×28 image | |
| ## Files | |
| Recommended structure: | |
| . | |
| ├─ app.py | |
| ├─ best_mnist_cnn.keras | |
| ├─ requirements.txt | |
| └─ README.md | |
| perl | |
| Code kopieren | |
| ## How to run locally | |
| ```bash | |
| pip install -r requirements.txt | |
| streamlit run app.py | |
| Notes | |
| For best predictions, upload an image with one digit, centered, with a clean background. |