YoloV5 test result metrics for single class

[djolodjolo23]

HI,

I have been using transfer learning yolov5 model on Edge Impulse, that I retrained on my dataset to detect the queen bee on a bee panel. However, the problem I am experiencing for the last few days is that I cannot interpret the testing results I receive.

calculate_object_detection_metrics {"support": 462, "coco_map": 0.6772876977920532, "coco_metrics": {"MaP": 0.032348874726968846, "MaP@[IoU=50]": 0.11738494455047443, "MaP@[IoU=75]": 0.005981072936587098, "MaP@[area=small]": 0.032348874726968846, "MaP@[area=medium]": -1.0, "MaP@[area=large]": -1.0, "Recall@[max_detections=1]": 0.10751445086705202, "Recall@[max_detections=10]": 0.10751445086705202, "Recall@[max_detections=100]": 0.10751445086705202, "Recall@[area=small]": 0.10751445086705202, "Recall@[area=medium]": -1.0, "Recall@[area=large]": -1.0, "support": {"images": 462, "annotations": 462, "detections": 350}}, "class_names": ["queen"]}

There are two different MaP values, the coco_map for me is more realistic because this model works well on unseen data, recorded in a very similar setup though. Any help is highly appreciated.