So, after trying out a lot of off the shelf models for detecting whether my baby was in or out of their crib, I decided that I’d probably need to just train my own. This allows the final model to be fairly small, and since it is fine-tuned on my particular crib/blankets/etc. it is a lot more accurate than stock yolo.
It connects to MQTT and publishes out auto-discovery topics that adds a binary sensor to Home Assistant.