Hybrid Object Detection Models

This repository contains two Python scripts demonstrating the use of hybrid object detection models combining YOLOv5 and Faster R-CNN. These models are fine-tuned to detect vehicles but can be adapted for other object detection tasks.

Overview

1. initial.py

   • Implements a pipeline where YOLOv5 is used for initial object detection, followed by Faster R-CNN for enhanced detection within the cropped regions identified by YOLO.
   • Focuses on processing images from a custom dataset and visualizing the results.

2. main.py

   • Demonstrates a more integrated hybrid approach combining YOLOv5 and Faster R-CNN.
   • Includes:
     • IoU-based merging of detections from both models.
     • Application of Non-Maximum Suppression (NMS) to refine final detections.
     • Visualization of results from YOLO, Faster R-CNN, and the hybrid model side by side.

Requirements

• Python 3.7 or above
• Required libraries:
  • torch
  • ultralytics
  • torchvision
  • opencv-python

Install the required libraries using:

pip install torch ultralytics torchvision opencv-python

Usage

1. Setup:

   • Ensure you have the pretrained YOLOv5 model file (yolov5su.pt) and a custom dataset of images.
   • Update the dataset_dir path in initial.py and the image_path in main.py to point to your dataset.

2. Run Scripts:

   • Execute initial.py for a basic detection pipeline:

     python initial.py

   • Execute main.py for hybrid detection with IoU-based merging and NMS:

     python main.py

3. Visualization:

   • View the results in separate windows displaying outputs from YOLO, Faster R-CNN, and the hybrid model.

Key Features

• YOLOv5:
  • Fast, real-time object detection with bounding boxes and confidence scores.
• Faster R-CNN:
  • Accurate region-based detection for refining YOLO results.
• Hybrid Approach:
  • Combines the strengths of both models using IoU for merging and NMS for final detection refinement.

Notes

• Ensure the image files are properly loaded; paths should be valid.
• You may adjust the confidence threshold and IoU values in main.py for optimal results.

Acknowledgements

• YOLOv5 by Ultralytics.
• Faster R-CNN from PyTorch's torchvision library.

Feel free to modify the scripts to suit your use case!