Clean Dataset & Train YOLO

I have a small but urgent computer-vision task. My goal is to detect red and green traffic lights on a Raspberry Pi robot, and I need a fully working YOLO model fast. I began with YOLOv11 in Google Colab, yet training fails because many label files are missing or incomplete. Images and TXT files are out of sync, and the folder hierarchy is a mess.

What I need from you:
• Re-organise the dataset into the standard YOLO structure (images/train, labels/train, etc.), rename files so every image has the correct label, and fix any annotation gaps.
• Train a model in Colab—feel free to use YOLOv5, v8, or stay with v11, whichever gives the best result quickly.
• Hand over best.pt (or equivalent), a Colab notebook that runs end-to-end, and the cleaned dataset zip.
• Provide a short set of commands or script so I can run real-time inference on my Raspberry Pi (Python, OpenCV, torch-vision).

Acceptance is simple: the notebook must train without errors and the exported weights must correctly flag red vs. green lights on my validation video. I’m on a tight timeline, so please be ready to start right away and iterate rapidly.