[Retracted] A Novel Attention-Based Lightweight Network for Multiscale Object Detection in Underwater Images

Table 4

Comparison of the model size and real-time performance of different methods.

Method	Backbone	Input size	Parameters ()	Model size (MB)	FLOPs ()	FPS
Yolov5	Darknet53		5.4	22.8	40.57	72
RON	VGG16		31.9	128.6	15.47	15
RefineDet	VGG16		50.5	200.2	18.98	40.7
STDN	DenseNet169		29.5	120.1	3.41	28.6
SWIPENet	VGG16		~	~	~	30
Faster R-CNN-AON	VGG16		84.1	336.3	23.67	24
RFBNet	VGG16		34.5	140.0	45.42	83
Ours	MobileNetv3		7.8	31.2	21.70	44.3