Table 6.

Summary of training hyper parameters used on different models for DDroneC-DB2 dataset.

	lightDenseYOLO (Ours)	YOLO v2	Faster R-CNN	MobileNets-SSD
Input size (unit: px)	Multi-scale training (from 128 × 128 to 640 × 640)	Multi-scale training (from 128 × 128 to 640 × 640)	320 × 320	320 × 320
Number of epochs	60	60	60	60
Batch size	64	64	64	64
Initial learning rate	0.0001	0.0001	0.0001	0.004
Momentum	0.9	0.9	0.9	0.9
Decay	0.0005	0.0005	0.0005	0.9
Backbone architecture	lightDenseNet	Darknet-19 448 × 448	VGG 16	MobileNets