This project provides an interactive 28x28 drawing grid, perfectly matching the MNIST dataset's input dimensions. It allows developers and enthusiasts to draw digits and receive real-time predictions from their own custom-trained TensorFlow or PyTorch models, offering a tangible way to assess model performance.
A custom-built canvas allows for intuitive drawing and erasing, with a blur effect to simulate natural handwriting and improve model accuracy.
Get instant predictions and confidence scores as you draw. The tool also includes a visualizer to display neuron activation values in real-time.
Supports both TensorFlow and PyTorch models with a flexible backend system. Custom conversion functions allow for compatibility with various model input formats.
The entire tool is available as a simple-to-install Pip package, allowing for quick integration into any Python project requiring interactive model testing.