Yogesh-ai3/Road-Damage-Detection ? reverse-engineered prompt

Build me a road damage detection project that can train on the RDD2022 dataset and find five kinds of surface damage in road images, including cracks and potholes.

I want the app to use a YOLOv8 model with transfer learning, then run inference on test images and save prediction text files in the right format for submission. Please make it easy to reproduce with a notebook for training and simple scripts for training and prediction too.

The dataset labels should follow the usual YOLO format, and the class names should match the five damage types from the README. After prediction, it should collect all output files into a submission zip folder so I can upload it for the competition.

Keep it practical for someone running it in Colab or on a modest GPU, and make sure the steps are clear enough that I can open the notebook and run everything in order. If you need to check the latest YOLO docs, feel free to look them up online.