Product Image Classification Model

I need an AI solution that takes product photos and sorts each image into fewer than five preset categories. The sole goal here is image classification—no detection or enhancement features are required.

You will start from an existing dataset of labelled product images (I’ll supply a download link once we begin). A lightweight, production-ready model built in Python with either TensorFlow or PyTorch is preferred; transfer-learning from a well-known backbone (e.g., ResNet, EfficientNet, MobileNet) is perfectly acceptable so long as the final classifier is accurate and quick to infer on standard CPU hardware.

Key deliverables:
• Clean, commented source code for training and inference
• The trained model weights (or exported SavedModel / .pt file)
• A short README explaining environment setup, how to retrain with new data, and a one-line CLI or REST example for running predictions
• Basic metrics report: training/validation accuracy, confusion matrix, and any augmentation techniques applied

Acceptance criteria:
• ≥90 % overall validation accuracy across all categories
• Clear, reproducible results when I run the provided script on a fresh machine

If you have questions about the dataset format or need additional samples, let me know early so I can provide them. Once the model meets the above criteria I’ll sign off on the milestone and we can wrap up promptly.