Table 3.

Single fish behavior detect performance comparison.

Method	Precision	Recall	F1-score	mAP50 (Detector)	FPS	Num. of parameters (Detector /Classifier)
Yolov4 + 3DCNN	0.97	0.96	0.97	0.78	38.59	27.6 M/ 4.2 M
Yolov4 + C3D	0.98	0.98	0.98	0.78	38.21	27.6 M/4.8 M
Yolov4 + Lite3D	0.98	0.99	0.99	0.78	38.87	27.6 M/0.08 M
Faster-RCNN + DCG-DTW	0.84	0.80	0.82	0.50	27.26	6.5 M/NA
Yolov4 + DCG -DTW	0.79	0.56	0.66	0.35	17.38	27.6 M/NA
EfficientDet-D1 + DCG -DTW	0.55	0.61	0.58	0.41	17.13	6.6 M/NA
BCS-YOLOv5	0.93	0.93	0.93	0.97	55	9 M/NA