How we use Machine learning at DigitalBridge

It’s difficult to renovate a bathroom. There are a thousand things to do, all of which a typical customer has never done before.

This problem is compounded by the imagination gap. When a customer views a product, it can be difficult for them to picture how that product will look in their bathroom. Is there a good place for that product? Is the product an appropriate size for the room? What does it look like in that space? These questions, and questions like these, are difficult to answer for a novice bathroom designer.

The mission of DigitalBridge is to guide a user through the design process, and answer questions such as these with ease.

We have examined and understood the customer journey from start to finish and identified points of friction that a typical customer faces.

The user journey a customer experiences when renovating their bathroom.

After an initial trigger that inspired a customer to renovate their bathroom, there are three challenging stages: inspiration, room capture and design.  

Let’s discuss each of these phases in turn, and the machine learning based solutions DigitalBridge offers to improve a customer’s experience.

Inspiration

An expert in interior design has a good grasp on the entire range of available products and different styles. However, a novice designer might not know what is available, let alone what looks nice together.

We have found that customers typically explore different bathroom designs through image searches. We call this search the Inspiration phase. A customer can view thousands of different designs through websites such as Pinterest and Google Image. However, when a customer views products from a retailer, they are presented with a list of different items. Current e-commerce solutions that retailers deploy do not account for the stylistic preference of a customer. As a result, there is a large gap between how a customer selects a product to purchase, and how a customer interacts with a retailer.

At DigitalBridge, we remove this source of friction in the inspiration phase of the customer journey through a novel suite of machine learning products.  

Customers’ style preferences are understood using a deep learning approach; a customer can upload a set of images reflecting bathrooms they like. DigitalBridge technology can identify the relevant products contained within the uploaded image, and compare these products to a vendor’s catalogue for similar items.

Before the customer interacts with the online services DigitalBridge facilitate, a vendor’s catalogue is processed. Each item in a vendor’s catalogue is analysed, and a style descriptor created. When a new customer uploads the image of a bathroom they like, the stylistic similarity between the products contained in the image and the catalogue products is computed.  Products that have a similar style to a customer’s preference can then be recommended. This allows users to interact with a retailer’s catalogue in a natural way.

An illustration of the style matching service provided by DigitalBridge.

Room Capture

During their journey to an ideal bathroom, customers are faced with the daunting challenge of measuring their current bathroom. Accurately measuring a bathroom is difficult and time consuming. Many customers do not even own a tape measure!

Whilst not every customer has a tape measure, most do have a smart phone. We have created a suite of machine learning products that run on a smart phone to automatically estimate the floor plan of a room.

DigitalBridge has taken two approaches to the problem of floor plan estimation. Our two approaches use different technologies. One solution to floor plan estimation utilises the 2D camera found on all smart phones. Our second solution uses the 3D cameras expected to be available in the next generation of smartphones.

Using the current generation of smart phones, which capture images and phone orientation information, DigitalBridge has constructed a machine learning based floor plan estimation algorithm.

A customer scans their room with their smart phone. Computer vision is used to analyse walls, and extract relevant information from each image. The floor plan can then be estimated using this information.

It takes a user around 4 minutes to scan their room. During the scan the floor plan can be evaluated with DigitalBridge’s algorithm to produce a floor plan accurate to centimetre precision. DigitalBridge’s technology is more accurate than currently available solutions that rely solely on a phone’s visual inertial systems, such as TapMeasure.  

The next generation of smart phones are expected to have 3D cameras. These next generation cameras capture a dense point-cloud, as well as a set of 2D images.

The point cloud captured of a kitchen captured by a 3D camera.

The point cloud constructed from video captured by a 2D camera.

To bridge the gap between current and next generation technology, a novel method of converting a video into a 3D point cloud has been developed.  Equipping customers with next generation technology, today.

We are forward thinking with technology, and ready for the change over to 3D cameras when it happens. In fact, we have already described how we can estimate an accurate floor plan from a 3D point cloud here. Take a look!

Design

With an accurate floor plan and appropriate pieces of furniture,  the next step in a customer journey is to design the bathroom.

Placing objects in a room is easy, but designing a bathroom is hard. What are the legal requirements on placing a sink? Just because a room can fit a very large sink, should it? These are questions customers may find themselves asking as they design their ideal bathroom.

DigitalBridge has constructed an algorithm to design elegant bathrooms which are always up to code. We have combined the legal requirements and common good practices to construct an auto-layout engine.  If you want to know more, take a look at this blog post from David!

The Catalyst auto-layout engine in action!

Utilising the big data sets available from customers using DigitalBridge products, we were able to learn what users consider good design. The implicit definition of good bathroom design used by customers was then augmented and validated with world leading designer input. The result is an auto-layout engine that removes the headache of bathroom design.

Conclusion

Through consistent innovation, DigitalBridge’s Machine Learning provides solutions to friction points in a customer journey. We’re utilising cutting edge technology to solve a real-world commercial issue, and enabling people like you or me to go back to what we really want to be doing in our spare time!

Follow us