Basic Usage of CLIP-GradCAM

This repo implements a GradCAM method based on the CLIP model. It is inspired by the official code: CLIP-GradCAM Example.

Prerequisites

Before running the code, ensure you have the following Python libraries installed:

• opencv-python
• numpy
• torch
• pytorch-grad-cam

How to Use

Adjust the parameters as needed:

• image-path: Specify the path to the input image.
• labels: Provide a list of descriptions for the image.
• prompt: Select one description from the list to use as the prompt.
• method: Specify the method for generating the class activation map. Supported methods include gradcam, gradcam++, scorecam, and others.

Example Command

Run the script with the following command:

python main.py --image-path "cat_and_dog.jpg" --labels ["a cat", "a dog"] --prompt "a cat" --method "gradcam"