Best way to calculate a neuron size of a model?

I am implementing transfer learning model from NVIDIA SoTA model: GitHub - NVIDIA/semantic-segmentation: Nvidia Semantic Segmentation monorepo

This model was learned from cityscape original image, the resolution is (1024 * 2048). The quality is extraordinary for me. So I need to calculate GPU memory that the model demands. I just write print(model) and calculate neurons in each layer manually. Holy moly! I know that task is nightmare. Is there a smart way to do?