Basic conv+ReLU+pooling layer extract the feature maps of an image.

• It passes through "Softmax" to determine whether anchors are positive or negative.
• Use the bounding box regression to fix the anchors in order to get the precise proposals

Collects the input of the feature maps and proposals. Combines those messages and extract the proposal feature maps

Calculate the class of the proposal by the proposal feature maps, and use the bounding box regression again to get the final precise point.

Recognition Rate