Business case

This model can be used, for example, to detect the appearance of a mass of bees at the mouth of a bee hive, which happens when one colony attacks another. The beekeeper must intervene immediately in such case.
The above-mentioned use can be achieved by examining the size of the bounding rectangles. While for individual bees the network detects small bounding rectangles, this model will only return a single large rectangle in the case of multiple insects next to each other.
This model detects all kinds of insects, so it can be used in other ways besides the bee detection mentioned above.

Number of classes

1 class: (insect)

Metrics

detections_count = 1868, unique_truth_count = 927
class_id = 0, name = Insect, ap = 86.34% (TP = 779, FP = 147)
for conf_thresh = 0.25, precision = 0.84, recall = 0.84, F1-score = 0.84
for conf_thresh = 0.25, TP = 779, FP = 147, FN = 148, average IoU = 71.17 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.863361, or 86.34 %

System requirements

Inference time using CPU: 300 ms (on HP Laptop 15-DA0042NH (Processor: Intel(R) Core(TM) i7-8550U CPU))