Grad-CAM Visualization Project

This project implements Grad-CAM (Gradient-weighted Class Activation Mapping) using TensorFlow and Keras with the pre-trained VGG16 model to visualize class-specific areas of an image that the model focuses on during classification.

Project Structure

├── gradcam.py           # Contains the Grad-CAM implementation
├── main.py              # Main app entry point
├── setup.py             # Your script to build the entire app
├── LICENCE.py           # The Licence that this code is distributed under.
├── predict_api.py       # API for handling prediction requests
├── POC.ipynb            # Proof of concept notebook
├── view/                # Folder containing UI interface for the app
├── artifacts/           # Folder to store input images and output Grad-CAM visualizations
│   ├── cat_dog.jpeg     # Sample input image
│   └── gradcam_output.jpg  # Generated Grad-CAM image
└── README.md            # This readme file

Requirements

The project relies on the following libraries:

• There is a requirements.txt file that have all you need

• tensorflow

• opencv-python

• numpy

• matplotlib

To install the dependencies, you can run:

You can use the setup.py file and it will do all the work for you via =>

Python setup.py develop 

pip install tensorflow opencv-python numpy matplotlib

How to Run

1. Download or prepare an image and save it in the artifacts/ folder (e.g., cat_dog.jpeg).

2. Run the main script to generate the Grad-CAM visualization:

python main.py

3. The script will display the original image along with the Grad-CAM heatmap, and it will save the output in the artifacts/ folder as gradcam_output.jpg.

Key Components

• Grad-CAM Class:
  The GradCAM class is used to calculate the heatmap of the areas in the image that contributed most to the predicted class.

• Preprocessing and Prediction:
  The preprocess_image function prepares the image for prediction using VGG16's expected input format. The model then predicts the class of the image using model.predict().

• Grad-CAM Heatmap Generation:
  The generate_and_save_gradcam function generates the Grad-CAM heatmap for the input image and overlays it on the original image.

Example Output

The example uses VGG16 to predict the class of an input image and visualize the Grad-CAM output. The model will output both the predicted class and a heatmap superimposed on the input image.

Sample Output:

Predicted: Rhodesian_ridgeback

A visual output (Grad-CAM heatmap) will be saved as gradcam_output.jpg in the artifacts/ folder and displayed using Matplotlib.

Additional Notes

• Change the Image:
  You can use any image by changing the image_path variable in the main() function.

• Modify the Model:
  The current implementation uses the VGG16 model. You can modify the model by loading a different one as long as it follows the same input format and has a convolutional layer for Grad-CAM visualization.

License

This project is licensed under the MIT License.