This project demonstrates the use of Convolutional Neural Networks (CNNs) in combination with Grad-CAM (Gradient-weighted Class Activation Mapping) for visualizing model interpretability in image classification tasks.
Grad-CAM provides visual explanations for decisions made by CNN-based models by producing a heatmap that highlights the most important regions in the input image contributing to the model’s prediction.
- ✅ Implemented a convolutional neural network using PyTorch
- 🧪 Trained the model on an image dataset
- 🧠 Applied Grad-CAM to visualize class-specific activation maps
- 🌡️ Generated heatmaps showing which regions most influenced the classification decision
The final output is a heatmap image overlaid on the original input image. This allows for:
- Better model explainability
- Understanding of what the CNN is “looking at”
- Debugging and verifying model behavior
- Detecting bias or irrelevant focus