This project provides a Grad-CAM visualization tool for YOLOv5 object detection models, enabling users to understand which parts of an image the model focuses on when detecting objects. It is targeted at researchers and developers working with YOLOv5 who need to interpret model behavior.

How It Works

The implementation integrates Grad-CAM, a technique for visualizing the decision-making process of convolutional neural networks, with the YOLOv5 architecture. It leverages the YOLOv5 codebase for model loading and the gradcam_plus_plus-pytorch repository for Grad-CAM computation, allowing for heatmaps to be overlaid on input images, highlighting areas of interest for object detection.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: YOLOv5 weights (auto-downloaded if not present), Python.
• Usage: python main.py --model-path yolov5s.pt --img-path images/cat-dog.jpg --output-dir outputs
• More info: python main.py -h

Highlighted Details

• Visualizes attention areas of YOLOv5 models.
• Compatible with YOLOv5-v6.1.
• Supports custom model weights and class names.

Maintenance & Community

• The project is maintained by pooya-mohammadi.
• References include gradcam_plus_plus-pytorch and ultralytics/yolov5.
• A citation is provided for research use.

Licensing & Compatibility

• The repository does not explicitly state a license in the provided README.

Limitations & Caveats

• The README notes an issue where truck heatmaps sometimes show nothing, which was reportedly solved in version 6.1.
• Future enhancements like GradCam++ and ScoreCam are listed as "To Do."