CNN output for Grid Sensor

I would like to learn more about the output of the CNN after feeding the "height x width x channel" matrix outputted by Grid Sensor. If my understanding is correct, this CNN is act in an unsupervised way for feature extraction, and the output should be like some clusters similar to the output of K-Means (please correct me if I'm wrong). After digging into the encoders.py script, I have figured out that the network structure is two conv2d layers, one flatten layer and one dense layer sequentially. However, I cannot figure out what the output of the dense layer is. Although from the utils.py script, I know that the output should be dimensionally equal to the h_size (number of hidden units per layer), I still cannot find any parameter related to this from the source code of Grid Sensor. Therefore, I was wondering if anyone could help me gain a better understanding of what is the output of the CNN?

Thanks!