Retail Object Detection with RF-DETR

Summary

This guide builds an automated beverage shelf-monitoring system in Roboflow Workflows: an RF-DETR model detects and counts Coca-Cola, Fanta, and Sprite bottles at the shelf front, a Custom Python block compares those counts to configurable targets and ranks products by replenishment priority, and Gemini 2.5 Pro writes a short inspection summary. The output is a single annotated image with detections, a shelf summary, and the highest-priority restock action, so staff know which shelf to fix first instead of walking every aisle.

Retail shelves change throughout the day as customers remove products, employees replenish stock, and bottles are moved between positions. Around 70% of retail out-of-stock events are linked to in-store execution, including delayed shelf replenishment, misplaced products, and inventory that remains in the backroom. Detecting low product presence early gives store teams a practical signal that a shelf needs attention before availability becomes a larger issue.

Most retailers still depend on employees to walk through aisles and inspect shelves manually. These checks take time and only show the shelf condition at the moment of inspection. When several products are running low, staff must also decide which shelf should be replenished first. Raw product detections alone do not provide that prioritization.

This project builds an automated beverage shelf-monitoring workflow using Roboflow workflows. An RF-DETR model detects Coca-Cola, Fanta, and Sprite bottles in shelf images, and the workflow counts the visible bottles in the monitored frame for each class. A Custom Python block compares the visible bottle counts with configurable shelf targets, calculates the shortages, and ranks products by replenishment priority. Gemini 2.5 Pro then analyzes the annotated image and generates a concise shelf inspection summary.

The system can detect and report:

Coca-Cola, Fanta, and Sprite bottles visible at the front of the shelf

The number of vis…