Shelf Item Counting POC

I need a compact proof-of-concept that can look at a single photo of a store shelf and return one simple number: the total quantity of products it finds. No bounding boxes, no section breakdowns—just the grand total per image displayed on a lightweight web dashboard. You will receive a small batch of sample shelf photos to start with. I’m open to whichever computer-vision stack you prefer (Python, OpenCV, TensorFlow, PyTorch, YOLO, etc.) as long as the final flow is clear and reproducible: • I upload or drag-and-drop an image. • Your model processes it and sends back the count. • The dashboard records and displays that number next to the image thumbnail. Accuracy does not need to be retail-grade yet, but the proof must show that the pipeline works end-to-end and can be fine-tuned later for brand recognition or section-level analytics if we extend the project. Deliverables: 1. Source code with brief setup notes (virtualenv or Docker preferred). 2. Running demo (local or small cloud instance). 3. Short read-me summarising model choice, training approach, and areas for improvement. I’ll test using a separate set of photos; if the dashboard reports a reasonable count within a few seconds per image, the POC is accepted.