How does CBIR (content based image recognition) fit into the problem of object detection? Let's say we want to detect 4 types of dogs (Golden Retriever, Cocker Spaniel, Greyhound, and Labrador). We have an "average" model trained using YOLOv3. So it might, for example, have a lot of false positives and false negatives.
How could we use CBIR to improve the detections from this "average' YOLOv3 model?