deep learning - Explain concept of instantiating a model with another model as argument? VGGFace(model='resnet50', include_top=False)

Question

I came across this line of code:

VGGFace(model='resnet50', include_top=False)

Could someone please explain to what this means? From my knowledge, VGGFace is a model trained to recognise faces, and then it accepts another model as an argument. So do we have two models? I am confused.

Thanks in advance.

question from:https://stackoverflow.com/questions/65942486/explain-concept-of-instantiating-a-model-with-another-model-as-argument-vggface

Answer 1

In fact it does recognizes faces, but the model parameter is used to specify which architecture is used, in this case resnet50, see: https://github.com/rcmalli/keras-vggface#available-models.

Don't know much about resnet50, but it's a residual network and that means that one of the earliest layers (usually the input) is feed back into the network in a later layer as a technique to increase accuracy on big networks (this one has ~150 layers).

More details: https://arxiv.org/abs/1512.03385