# intersection over union

A metric commonly used in Object detection. In this task, an image is labeled with ground truth bounding boxes. The model predicts bounding boxes. The IOU is how much a predicted box overlaps with the bounding box, divided by the union of their two areas.

Then, we can set a threshhold on the IOU that we consider a "hit". So for a ground truth bounding box and a predicted bounding box, we consider the model to have "hit" the ground truth if it is above the threshhold. We can then compute precision and recall (see Precision vs Recall) depending on whether the "hits" are true positives, false positives, etc.