PyTorch Implmentation of Grad-CAM.
Grad-CAM is a technique to generate visual explaination of the result of a Convolutional Neural Net.

• gradcam.py contains the codebase for the Grad-CAM method. The GradCam class expects a model and its layer as input.
• test.py is a sample script to generate gradcam output.

Above Implmentation test uses Resnet101 model. But any model can be used to generate output.

See the Paper: Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, Dhruv Batra; Grad-CAM: Visual Explanations From Deep Networks via Gradient-Based Localization.

Authors' implementation: ramprs/grad-cam.