Herman Miller, a 100-year-old-plus company, is an industry leader in office and home furniture, which are sold and used in countries all over the world.

Herman Miller furniture is highly customizable, with thousands of combinations for each piece including choices for the color and pattern of the fabric. In addition, customers can not only choose from a catalog of Herman Miller fabrics, they can also request a fabric of their own, making an order even more complex.

When a custom fabric is requested, the process of manually searching for a similar fabric in Herman Miller’s existing catalog of fabrics is tedious, error prone and very time-consuming.

Our Fabric Identification Based Recommendation Engine, FIBRE, leverages computer vision and machine learning to classify fabrics, automatically detecting color and pattern.

FIBRE is first applied to Herman Miller’s existing extensive catalog of fabrics, to tag images with standard, quantifiable measures of both color and pattern.

When a request for a custom fabric is submitted, our system analyzes it, generates its measures of color and pattern, searches the extensive Herman Miller catalog of fabrics, and finds fabrics that most closely match. The four most similar fabrics are displayed for review by a Herman Miller associate or customer, thereby reducing the number of fabrics to consider from thousands to just four.

Our FIBRE utilizes scikit-learn for color detection, and SageMaker and TensorFlow for pattern detection. Flask provides the client side interface to the backend, which is hosted on Amazon Web Services.