IoU Calculation over a set of images and several predicted anchor boxes

Question

I have a set of images with one ground truth boxe per image and six classes in total. I'm only interested in predicting the boxes, and not so much about the classes themselves.

After my prediction, I got several anchor boxes per image. To calculate IoU, I compared the ground truth per image with each predicted box, per image. My question here is: how am I supposed to calculate a total IoU for the model? I assume I should calculate a mean value per image, and then calculate also a mean for the total model, based on the mean IoU value per image. Since I'm not quite sure about this assumption, I wanted to double check if that would be the right approach to go.

Thanks for any help in advance