Volumental produces shoe recommendations for millions of shoppers by leveraging a combination of 3D foot scans, retail purchase data, and AI. Their team partnered with Sama to label the datasets that fuel the computer vision technology for their mobile foot scanning app — one piece of Volumental’s technology suite which empowers retailers and brands to create frictionless and more personalized experiences for their customers, both in-store and online.
Ordering shoes online can feel like a gamble. How many of us have anticipated the arrival of a fresh pair of sneakers, only to have to promptly return them due to poor fit? According to one study, about 52% of us have experienced that disappointment.
Shoe size labels can be a poor indicator of actual fit, both across categories of footwear and even within the same brand. And with an estimated $428 billion in merchandise returned to retailers in 2020, there exists a huge opportunity for retailers and brands to do some damage control.
For Sweden-based Volumental, the answer isn’t to solve for consistency or standardization within the shoe manufacturing process, but rather to leverage AI to deliver more accurate fit recommendations: an industry segment they are dubbing FitTech™.
AI can help deliver more delightful, personalized retail experiences
For many years, Volumental has been outfitting shoe retailers with in-store scanners which capture 3D scans of customers’ feet within seconds. These scans are cross-referenced with their extensive database of retail purchase data to deliver product recommendations to shoppers, based on accurate foot measurements and the purchase behavior of consumers with similar feet.
This technology empowers shoppers to make good buying decisions, but it also supplies retailers with data-driven insights to help them provide more personalized recommendations to consumers both in-store and online.
Armed with millions of foot scans and a database of purchase behavior from customers of nearly 100 of the world’s top retailers and brands, Volumental then set out to tackle the challenge plaguing e-commerce stores and shoe-wearers everywhere: online returns due to poor fit.
(Image source: Volumental website.)
The Volumental mobile app delivers the same benefits to consumers, but this time, from the comfort of their homes. The user experience is seamless: take a few photos of your feet from key angles and receive accurate foot measurements along with data-backed recommendations for shoes that are sure to fit like a glove.
Data labeling challenges for high-precision foot scanning
The Volumental mobile app user experience may be straightforward, but the challenge of developing this technology was not. To deliver a seamless experience in the app, Volumental had to solve a range of technical problems.
While the LiDAR capabilities that come equipped in modern-day smartphones work well to make many AR experiences more accurate and realistic, they are not useful for foot scanning. Existing AR frameworks on the market did not provide the level of accuracy required for Volumental’s mobile foot scanner, so they set out to build their own proprietary models.
These images of Volumental employees’ feet show the difference between foot scans reconstructed with sensors native to modern smartphones (left) vs the level of precision required to train their own proprietary models: pixel-perfect masks (right). (Source.)
To obtain the high-accuracy segmented images they needed to power their mobile scanning, Volumental began the search for a data labeling partner. They knew that the dataset had to adhere to an extremely high standard of quality if their mobile solution was to deliver the seamless experience their users were accustomed to in-store.
Mikael Andersson, Sr Product Owner at Volumental explained why this requirement was imperative:
“For our mobile app, we needed extremely precise segmented data because we knew that every missed pixel would easily add up to millimeters of lost accuracy.”
In addition to a high standard of label quality, the data labeling project would also require annotators to know how to handle edge cases such as shadows, low-contrast light, and occlusions. These exceptions needed to be handled with consistency if the training data was to make Volumental’s algorithms behave predictably. For these edge cases, and to ensure that their solution would be able to scale without compromising quality, Volumental needed tight feedback loops to keep their models high-performing.
To meet all these requirements, Volumental partnered with Sama to deliver high-precision segmented labels for their datasets. Mikael explains:
“Having worked with different cloud providers where the staff doing the actual work was always very hidden from us, we appreciated the transparency Sama gave us. They were communicative and very easy to work with from data collection to project management.”
Delivering a delightful and uncompromisingly accurate experience to their users was important to Volumental, but so was social sustainability. As CMO Brent Hollowell explained, building a mass-market experience that is representative of the world must include diverse and representative datasets:
“At the core of our interest in social sustainability are inclusivity and cultural diversity. Diversity of data is the strength of our AI-powered Fit Engine and it’s also what helps us succeed as a company.”
Partnering with Sama assured Volumental that the diversity of their datasets would be accurately represented in their models: to deliver hyper-personalized experiences to delight their users across the globe.
Better labels lead to better business outcomes
In part thanks to accurately labeled data, Volumental is creating more delightful omnichannel shopping experiences for consumers. Accurate 3D foot scans are just one piece of the puzzle: combined with their extensive database of purchase behavior and proprietary ML algorithms, Volumental can deliver hyper-personalized recommendations to consumers.
These recommendations don’t only provide better shopping experiences for users, they remove friction from the buying process and result in fewer online returns. The end result? Happy, loyal customers and ultimately, more revenue for retailers and brands — all thanks to AI-powered fit recommendations.