Tikam02/CrowdEstimation ? reverse-engineered prompt

Build me a crowd counting and crowd density estimation project that can look at an image and estimate how many people are in it. I want it to feel like a real deep learning demo, not just a notebook with a few random cells.

Please make it based on the ideas from the common crowd counting papers, especially CSRNet and the older multi column and multitask crowd counting approaches, and keep it organized so I can compare the different methods. Use the ShanghaiTech crowd dataset if needed, and set it up so the training, testing, and prediction steps are easy to run in notebooks.

I’d also like a simple way to show the predicted crowd density and trigger an alert if the count goes over a threshold, since that’s the main real world use case I care about. If you need to check current setup details or docs online, feel free to do that. Keep it practical, easy to understand, and ready to run in Jupyter.