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

Table 2

Comparison of AP values of different methods. The short names are defined as HO—holothurian, BF—big fish, DI—diver, CO—coral, RO—rock, SC—sculpture, SF—small fish, OC—octopus, TU—turtle, SW—seaweed, MR—manta ray, SF—starfish, SH—shell, SU—sea urchin, and WR—wreckage. The best results are marked in bold.

Method	mAP	HO	BF	DI	CO	RO	SC	SF	OC	TU	SW	MR	SF	SH	SU	WR
Yolov5	71.27	72.80	75.54	76.11	63.92	64.54	63.57	69.41	72.61	73.36	64.39	75.49	70.53	69.34	70.02	67.37
RON	74.26	71.93	78.32	77.29	65.47	67.18	68.95	72.87	74.49	75.02	69.71	78.32	73.68	74.33	74.16	70.12
RefineDet	75.06	69.17	77.09	78.31	68.71	67.34	70.13	76.29	76.38	75.85	70.24	78.37	74.92	75.06	75.13	73.21
STDN	76.67	73.28	80.14	77.42	69.18	67.99	69.71	78.51	77.89	76.91	74.06	77.91	75.03	74.91	76.01	74.88
SWIPENet	78.91	70.35	80.59	80.58	70.26	69.43	73.55	79.48	78.01	78.25	74.57	79.38	75.86	79.08	75.64	75.61
Faster R-CNN-AON	79.68	73.01	84.47	86.03	71.25	70.05	73.69	78.52	79.62	78.84	75.29	79.94	78.61	76.31	78.35	75.36
RFBNet	73.99	70.67	83.92	82.63	69.04	67.36	70.19	76.13	74.17	73.90	70.37	74.06	73.31	70.91	72.05	70.91
Ours	81.94	78.33	83.98	85.14	73.38	72.01	73.30	78.57	80.85	82.41	75.17	81.54	81.02	79.63	80.28	74.23