I have a collection of scanned images from physical product stickers and need a machine-learning solution that can reliably spot the product codes printed on them. The scans vary slightly in lighting and angle, but resolution is consistent. Here’s what I need: • A trained model (deep-learning OCR or a custom computer-vision pipeline with TensorFlow, PyTorch, OpenCV, or equivalent) that extracts product codes from each image and outputs them in a structured format such as CSV or JSON. • An inference script I can run locally to process new scans in batches. • Brief documentation that explains setup, dependencies, and how to retrain if I add more sticker samples. Acceptance will be based on: • 95 %+ recall on a held-out test set I provide. • Correct mapping between each filename and its detected code(s). • Clear, reproducible instructions. I’ll share an initial dataset of labeled scans once we start. Let me know which framework you prefer and any data requirements so I can prepare the files accordingly.