How to get the visual observation from Unity into Python?

Hi, I saw the observation.proto file in the protobuf-definitions.

May I know how to retrieve and load the image from the camera sensor/visual observations before the image is fed into the trainer/neural network?

Also, may i know which file contains the portion of code where the camera sensor image from Unity is fed into the network as input for training?

Thanks! Am a beginner programmer and there are so many files that made me lost.

you can take a look at network.py file