Product that will recognise building quality
Budget: $250 – $750 USD
I am building an AI-driven workflow that can automatically look at a product photo and assign it to the correct category. The end goal is to drop an image into the system and instantly receive the predicted class label so that my catalogue stays organised without manual tagging.
The scope centers on image recognition—specifically image classification—so the core deliverable is a production-ready model trained on product images. I have an initial set of labelled pictures, but I am open to your advice on augmenting or cleaning the dataset if that will push accuracy higher. You may work in TensorFlow, PyTorch or another modern framework as long as the final solution is portable and easy to retrain when new categories appear.
Acceptance criteria
• Model reaches an agreed-upon accuracy on a held-out validation set
• Inference script or lightweight REST API that takes a single image and returns the predicted class in JSON
• Clear README with setup steps and pointers on how to continue training with additional data
Please include a brief outline of your preferred approach, any relevant past projects in computer vision, and a rough estimate of training time once data is in place.
The scope centers on image recognition—specifically image classification—so the core deliverable is a production-ready model trained on product images. I have an initial set of labelled pictures, but I am open to your advice on augmenting or cleaning the dataset if that will push accuracy higher. You may work in TensorFlow, PyTorch or another modern framework as long as the final solution is portable and easy to retrain when new categories appear.
Acceptance criteria
• Model reaches an agreed-upon accuracy on a held-out validation set
• Inference script or lightweight REST API that takes a single image and returns the predicted class in JSON
• Clear README with setup steps and pointers on how to continue training with additional data
Please include a brief outline of your preferred approach, any relevant past projects in computer vision, and a rough estimate of training time once data is in place.