iotaisolutions/BMWQuantumChallenge2021 ? reverse-engineered prompt

Build me a notebook based project that shows how to detect surface cracks in images and compare a few different ways of doing it. I want one version with a normal image classifier, one with a quantum neural network, and one that uses quantum kernel alignment to improve the quantum model. Use the Surface Crack Detection dataset from Kaggle, and make it work in a way that can run locally or in Colab.

Please include the full training and evaluation flow, plus clear notebook style explanations so someone can follow along. It should show the data loading, image preprocessing, model training, and accuracy comparison between the classical and quantum approaches. If needed, look up current docs online to make sure the Qiskit and PyTorch parts are set up properly.

Also add a simple report or summary of the approach, what each version does, and why the quantum kernel alignment version is useful for this problem.