Lightweight Object Detection Ensemble Framework for Autonomous Vehicles in Challenging Weather Conditions

Table 2

Best performing models for OD on the DAWN dataset with color augmentation.

Model	Voting strategy	Class 1 (pedestrian) AP (%)	Class 2 (bicycle) AP (%)	Class 3 (car) AP (%)	Class 4 (motorbike) AP (%)	Class 5 (bus) AP (%)	Class 6 (truck) AP (%)	mAP (%)
RetinaResnet50	Unanimous	50.53	17.19	49.71	25.93	21.73	13.18	29.71
RetinaResnet50	Affirmative	52.34	23.29	52.56	35.51	19.09	13.71	32.75
Yolov3 + RetinaResnet50	Affirmative	34.03	23.29	33.98	35.51	13.56	13.71	25.68
RetinaResnet50 + SSD	Affirmative	35.63	23.29	34.79	35.51	14.28	13.71	26.20
Yolov3 + RetinaResnet50 + SSD	Affirmative	32.53	23.29	32.51	35.51	12.59	13.71	25.03