| --- |
| license: mit |
| language: |
| - en |
| metrics: |
| - accuracy |
| pipeline_tag: image-classification |
| tags: |
| - art |
| --- |
| # Realtime Digit Draw |
|
|
| **Framework:** PyTorch |
| **Purpose:** Interactive handwritten digit recognition for learning and experimentation. |
|
|
| --- |
|
|
| ## Overview |
|
|
| Realtime Digit Draw allows users to **draw a digit in a Pygame window** and see a real-time prediction from a trained ANN model. |
| This project is **for learning purposes** and demonstrates a simple ANN trained on MNIST with live inference. |
|
|
| --- |
|
|
| ## Features |
|
|
| - Draw digits directly using the mouse |
| - Real-time predictions using a small PyTorch ANN model |
| - Preprocessing compatible with MNIST 28×28 grayscale inputs |
| - Lightweight and fully offline |
|
|
| --- |
|
|
| ## Files |
|
|
| | File | Description | |
| |------|-------------| |
| | `app.py` | Main interactive drawing UI | |
| | `model.py` | ANN architecture definition | |
| | `inference.py` | Load model and predict digits | |
| | `weights.pth` | Trained model weights | |
| | `train.py` | Script used for training the model | |
| | `requirements.txt` | Required Python packages | |
| | `.gitignore` | Folders/files ignored in the repo | |
