Python Deep Learning - Second Edition

Model architecture

The following is the ChauffeurNet model architecture:

(a) ChauffeurNet architecture and (b) the memory updates over the iterations (source: https://arxiv.org/abs/1812.03079)

First, we have FeatureNet (in the preceding diagram, (a)). Its inputs are the middle-level top-down images we defined in the Model inputs and outputs section. The output of FeatureNet is a feature vector, F, which represents the synthesized network understanding of the current environment. This vector serves as one of the inputs to the recurrent network AgentRNN. Let's say that we want to predict the next point of the agent's trajectory (step k). Then, ...