Accurate Detection and Tracking of Small-Scale Vehicles in High-Altitude Unmanned Aerial Vehicle Bird-View Imagery

Table 1

Main parameters adopted in the training experiment.

Parameters	Value and unit
Input size	768 × 768 pixel
Training epochs	300
Warm-up epochs	5
Number of classes	3
Batch size	16
Optimizer	SGD
Scheduler	Warm-up + cosine annealing
Initial learning rate	0
Minimum learning rate ratio	0.05
Momentum	0.9
Weight decay	5e − 4
Detection threshold	0.01
IoU threshold for softer NMS	0.65
No augmentation epochs	30