TABLE II. The Network Specifications for Regression Subnetwork: The Kernel Size and Stride, and the Output Size (Channel 
Height 
Width) of Each Layer. The Regression Network is Fed With the Concatenation of the Input Image and the Detection Output Maps, and Outputs Stacked (
) Probability Maps With the Same Size as the Input Image.
| Kernel (Size, Stride) | Output ( ) |
|
|---|---|---|
| CBR1 |  |
 |
| CBR2 |  |
 |
| CBR3 |  |
 |
| CBR4 |  |
 |
| CBR5 |  |
 |
| CB |  |
 |