Shelf Counting Vision POC

I need a quick proof of concept that can process photos I take of retail shelves with my phone, automatically count every visible item, and display that total in a lightweight web-based dashboard. Core requirements • Accept JPG or PNG uploads (mobile-phone resolution). • Run a computer-vision model—YOLO, Detectron, or any comparable open-source alternative is fine—to detect individual objects. • Return a single integer per image representing the total items found. • Present results in a clean dashboard table: image thumbnail, filename, and count. No per-section breakdown or on-image marking is necessary for this POC. Tech notes Python is preferred, using OpenCV, TensorFlow or PyTorch, but I’m open to efficient alternatives. A simple Flask, FastAPI, or Streamlit front end is enough; focus on functionality over design. Deliverables 1. Source code with clear setup instructions (README). 2. Pre-trained weights or training notebook if custom training is required. 3. Short demo video or live link proving it works on at least five sample shelf photos. 4. Brief report outlining model choice, accuracy observations, and next-step recommendations. Keep it lightweight and well-documented so I can extend it later to identify product types or highlight empty spaces.