An Approach to Improve SSD through Skip Connection of Multiscale Feature Maps

Table 2

Training parameter settings.

Parameter	Size
Input size	300 × 300
Number of iterations	240,000 steps
Batch size	16
Initial learning rate	0.0005 (it is divided by 10 in steps of 160,000, 200000, and 240,000)
Weight learning rate	0.0005
Momentum	0.9