k4ilash/image-interpretability-demo

Interpretability Methods Throughout the Years (with MNIST)

Draw a digit or load a random MNIST sample, select an interpretability method, and view the saliency overlay. This project follows this article and demonstrates various neural network interpretation techniques on handwritten digits.

Features

• —Interactive digit drawing canvas
• —Multiple interpretability methods (Vanilla Gradient, and others)
• —Real-time saliency visualization
• —Probability distribution display
• —Smooth overlay options

Usage

1. 1.Draw a digit on the canvas or load a random MNIST sample
2. 2.Select an interpretability method from the dropdown
3. 3.Choose target class (auto-detection or manual selection)
4. 4.View the saliency overlay showing what the model focuses on