I’m using the Ultralytics YOLOv8 model for object detection on my custom dataset (which has two classes). After running predictions, I want to calculate and save the Intersection Over Union (IoU) for each detected bounding box alongside the confidence scores.

Here’s the relevant part of my code where I run the predictions:

from ultralytics import YOLO
import os

# Initialize the YOLO model with pre-trained weights
model = YOLO('runs/detect/yolov8n/weights/best.pt')

# Directory containing test images
images_directory = 'path_to_test_images_directory'
results_directory = 'path_to_save_predict_results_yolov8n'

os.makedirs(results_directory, exist_ok=True) 

image_files = [os.path.join(images_directory, f) for f in os.listdir(images_directory) if f.endswith(('.jpg', '.png', '.jpeg'))]

for image_path in image_files:
    results = model.predict(source=image_path, save_txt=True, save_conf=True, show_labels=True, show_conf=True)
    
print("All images have been processed and results saved.")

This is the one of the sample of predicted boxes on image.

0 0.75847 0.525917 0.48199 0.516998 0.976422
0 0.223181 0.527131 0.446362 0.516285 0.973426
1 0.487469 0.288411 0.0540745 0.0805167 0.836555

I am able to save the predictions and confidence scores but am unsure how to calculate and store the IoU for each prediction relative to the ground truth bounding boxes, which I have available in a separate dataset.

Questions:

1. How can I modify my code to calculate the IoU for each predicted
   bounding box against its corresponding ground truth?
2. What would be an efficient way to save these IoU scores
   alongside the confidence scores for further analysis?

Any guidance or sample code would be greatly appreciated!