Train Lightweight Image Classification Model

I need a trained deep learning model that can recognize approximately ~60 types of fruits and vegetables from camera images in a retail store environment.Use case: The camera will be mounted above scales in a store. A customer or seller puts a package, net, or bag with produce on the scale. The model should quickly recognize the main type of vegetable/fruit inside (for example: "purple onion", "potato", "carrot", etc.).Requirements:Base dataset: Fruits-360 (or similar) + fine-tuning for the target ~60 classes The model must work in real-time (fast inference) Must handle packaged produce (bags, nets), not only clean single items Lightweight model preferred (suitable for edge devices or fast processing) Classification of the main/dominant object in the frame (not full object detection with multiple bounding boxes) Deliverables (must be provided):Trained model file in a convenient format for inference (.tflite is preferred, also .onnx or .pt) Python inference code / script (example how to run the model on a new image or camera frame) Jupyter Notebook or Python script for further fine-tuning / retraining on new data File with class names (label mapping) Short evaluation report (accuracy on validation set, etc.) Detailed README file with:How to use the model for inference Step-by-step instructions how to retrain/fine-tune the model on new photos Recommendations for data collection Additional info:Number of classes: ~60 (exact list will be provided after we start) No mobile app or user interface needed — only the model + code English class names are acceptable Please include in your proposal:Your suggested model architecture Estimated timeline Fixed price for the whole project Links to 1-2 similar computer vision projects you have done Budget: Fixed price (please propose your price)