The Challenge

Finding new ways to improve in-store customer service.

A leading athletic retailer wanted to improve their in-store customer experience. They approached Credera to develop a tool that would allow shoppers to identify products in-store and return additional product information to their mobile device without the need for a sales associate. By leveraging deep learning models and augmented reality technology,  Credera was able to reliably identify a shoe and return relevant information.​

The Solution

Turning machine learning into a customer benefit.

• Leveraged Google Cloud Platform to build a scaleable machine learning environment, utilizing docker to handle model and deployment dependencies​.

• Performed various image wrangling techniques such as resizing, cropping, and rotation to boost overall training set size and improve generalizability​.

• Utilized a two-model approach with Keras and Tensorflow 2.0 that utilized deep learning models YoloV3 and MobileNet2 to identify and classify each shoe, respectively​.

• Developed both an iOS and Android augmented reality application that integrated the models to classify shoes on a mobile device​.

• Architected a model training pipeline that would allow new products to be added to the model and continually trained upon.

The Results

Proving the value of augmented reality and machine learning.

RELIABLE PREDICTIONS

Predicted the product correctly approximately 85% of the time when tested against our proof of concept shoe set​.

CONTINUAL TRAINING PIPELINE ARCHITECTURE

Developed a model training pipeline that once operationalized would allow new products to be added to the model and dynamically trained on​.

MACHINE LEARNING BEST PRACTICES

Coached their team on how to maintain a machine learning model and how to harbor a quality machine learning environment​.

AUGMENTED REALITY ENABLEMENT

Demonstrated how augmented reality can be utilized with machine learning models and laid the foundation for further augmented reality development.