OpenCV has methods for importing models built with deep learning frameworks. We load the TensorFlow SSD model as follows:
config = "./ssd_mobilenet_v1_coco_2017_11_17.pbtxt.txt"model = "frozen_inference_graph.pb"detector = cv2.dnn.readNetFromTensorflow(model,config)
The first parameter of the readNetFromTensorflow method accepts a path to a file that contains a TensorFlow model in binary Protobuf (Protocol Buffers) format. The second parameter is optional. It is a path to a text file that contains a graph definition of the model, again in Protobuf format.
Surely, the model file itself might contain the graph definition and OpenCV can read that definition from the model file. But, with many networks, it might be ...
