Visual Explaination from Deep Neural Networks with CAM and Grad-CAM Techniques and Create Bounding-Boxes:

This Repository contains the implementation of visualization techniques for Deep Learning Neural Networks and also contains the implementation for creating Boundary-Boxes (Rectangle and Polygon) with these techniques. The visualization techniques are

• Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization

  • Creation of BoundingBoxes (Rectangle and Poygon) using a Grad-CAM Heatmap

• CAM: Learning Deep Features for Discriminative Localization

Code Implementation and Results:

Implementations:

Grad-CAM:

Paper

Code: Google Colab Notebook

Bounding-Boxes with GRAD-CAM:

The following notebook creates BoundaryBoxes (Rectangle and Polygon) using Grad-CAM Heatmap output.

Code: Google Colab Notebook

CAM:

Paper

Code: Google Colab Notebook

Results:

Synthetic Data:

Tuberculosis Data:

BRATS-2017 Data:

Bounding-Boxes with GRAD-CAM for Tuberculosis: