[Python] How to access the base train class in YOLOv11 from the Ultralytics GitHub repository?

I'm currently working on training a custom object detection model using YOLOv11, and I'm diving into the Ultralytics GitHub repository to better understand the internal structure of the training process.

I’ve noticed that older versions like YOLOv5 or YOLOv8 have well-defined training functions and sometimes even a train.py file with a clear entry point. However, in YOLOv11, I couldn’t locate a specific base train class or a modular training component that can be directly extended or overridden.

I’m trying to access the core logic that handles the training loop — such as optimizer setup, dataloader creation, forward pass, loss calculation, and backpropagation — so that I can integrate my own metaheuristic optimization (e.g., GWO, PSO) to dynamically adjust hyperparameters.

My questions are:

1. Does YOLOv11 have a base training class or script that encapsulates the training loop?
2. If so, where in the Ultralytics YOLOv11 repo can I find it?
3. Is the training process callable as a function or is it tied directly to CLI execution?
4. Any tips for customizing training or plugging in an external hyperparameter tuning loop?
5. Any help or guidance would be appreciated — especially if you've done similar modifications or worked with YOLOv11's internals.

Thanks in advance!

Continue reading...