Artificial Intelligence is revolutionizing the retail industry by enhancing efficiency, personalization, and customer satisfaction. AI-powered tools analyze vast amounts of data to predict consumer trends, optimize inventory management, and streamline supply chains, reducing costs and waste. Personalized shopping experiences are created through recommendation engines that leverage machine learning to suggest products based on customer preferences and browsing history. 

In this project I helped develop the ML pipelines for in store retail applications. Mounting cameras on shelves, training models to detect and identify SKUs, alert stock personnel to gaps in inventory on edge. 

---

Rearchitecting Yolo for Edge Inference

Deploying on edge accelerators come with unique challenge including the limited operation set available for the IMX. This requires the refining of layers to available operations, limiting output tensor dimensionality by altering number of defined anchors. 

​

​

Planogram and Gap Detection

Using Yolo we were able to determine with high recall the locations of objects on shelf. Cropping these detections and feeding them to an embedding system provides higher SKU identification than class probabilities from yolo alone. 

Patterns of Life

​ID obscured human detection allowed us to maintain heatmaps of traffic through the isles. Determining high traffic areas allows retailors to provide greater granularity in cost and sales projects for prospective suppliers.