Issue for training on faster rcnn

[FrankCAN]

Hi, thanks for your code, but I am quite confused for the difference between training on ssd and faster rcnn.

As we know, ssd is one stage detection model, the feature extraction part is already included in the model, there is no problem. But what I am confused is training on faster rcnn, because faster rcnn is two-stage training process, which need a feature extraction model (e.g. vgg16) in advance before training detection model. So I am very confused that how to use Object Detection API to train faster rcnn, because I didn't find any trained feature extraction model. Besides, do we need to train VGG16 model separately for our own dataset?

Many thanks for your help. Frank