Building Retail Experiences Using Machine Learning
Increasingly over the past couple of quarters, we have been working with more and more retail customers. A common theme has emerged during these discussions that centers on building the best customer experiences as they traverse through a brick and mortar store. The experiences retail environments want to innovate in include:
- Making the checkout process easy
- Managing assets efficiently
- Visualizing real-time analytics in the store
- Forecasting inventory in real-time
- Marking customer journeys in the store, etc.
Another top of mind question is, “can we use video to anonymously capture journeys, product trends, etc., while keeping privacy and security as a key requirement?”
In our DevNet Co-Creations Labs, we have been working on a Machine Learning innovation called FlowIQ. FlowIQ pertains to vector-based object recognition and more specifically to machine-learning based models developed in hybrid clouds on which vector searches may be performed for object recognition.
Developer tools for building machine learning models
Cisco recently announced the UCS C480 ML – a member of the UCS family of servers that includes 8 Nvidia v100 GPUs. It has an optimized design for AI/ML/DL (artificial intelligence, machine learning, and deep learning). In fact, our team recently even validated it for RAPIDS, making it extremely easy for enterprise ML developers to build ML models.
For the retail use case here, we used traditional tools to do our data gathering, cleaning and model building. Some of tools include MxNet, Tensorflow, OpenCV, Numpy and datasets include MS1M, VGG2, LFW, and MegaFace.
Real-time object detection and recognition at the edge
For the retail case, the application used video streams from IP cameras. The video gateway at the edge analyzed the video streams in real-time, did object detection and recognition at the edge (Nvidia Jetson TX2) and passed the object to the private data center. The recognized object, and all the associated meta-data, were stored at this layer.
Customer privacy is protected
As far as privacy is concerned, the application does not store any video or images at all; the only thing that is stored is a vector of the object and associated meta data like position and timestamp. The retail application that was built used the FlowIQ platform. This application is a simple Check-in/Check-out and journey mapping app using facial recognition and generating a dynamic QR code profile for privacy reasons. From the customer point of view, their privacy was protected as they were using a dynamic profile in the store.
From the retailer point of view, now they can map customer journeys anonymously. They can even embed this into their mobile retail apps that customers can use to make payments or they can send specific customers special offers or advertisements.
Unleash your power of innovation
This is just the beginning of unleashing the power of data by using machine learning. Machine learning will play an important part in building new solutions for your line-of-business, and Cisco DevNet is here to help. Check out the Cisco DevNet ML/AI Dev Center, where you’ll find resources for learning and building solutions with AI and ML.
Does this solution interest you? Visit Cisco DevNet’s Co-Creations and get in touch with us.
Want to get hands-on with new Cisco AI and ML technology? Read Chloe Kauffman’s blog on how the new UCS C480 ML helps you adapt to new requirements.
Get your free DevNet account for access to network programmability and application development resources, learning labs, and sandboxes.