Adaptive Multi-modal Fusion Instance Segmentation for CAEVs in Complex Conditions: Dataset, Framework and Verifications

Chinese Journal of Mechanical Engineering

Table 1 Comparison of single modal and different fusion operations on the Complex KITTI dataset

Input	Approach	AP	AP₅₀	AP₇₅
RGB	Single modal	24.74	39.76	27.16
DDM	Single modal	24.39	41.42	27.27
RGB + DDM	Sum fusion [44]	25.22	43.09	27.85
RGB + DDM	Maximum fusion [57]	25.69	43.02	28.83
RGB + DDM	Sharpening fusion [19]	26.97	45.55	29.21
RGB + DDM	MoE + Sum fusion [12]	26.96	44.74	29.50
RGB + DDM	MoE + Maximum fusion	27.54	45.69	30.65
RGB + DDM	MoE + Sharpening fusion (ours)	29.76	49.36	32.07