How Perfitt Uses Labeling Automation to Help Customers Find the Perfect Footwear Fast
Perfitt’s technology requires a high level of accuracy to successfully match a user’s foot to the proper shoe size across brands. They needed to create an efficient data pipeline from data collection to model training to continually improve model performance.
Superb AI’s custom auto-labeling capabilities, along with advanced data management and analytics, work status indicators, and QA tools.
90% of raw and labeled data managed in one place
1.5x faster polygon segmentation with custom auto-label
Active learning workflows established
Reduced processing time through increased data quality
Perfitt, a Korean startup, is transforming the online shoe shopping experience by employing artificial intelligence to find the perfect fit for customers. This innovative approach eliminates the need for physically trying on shoes. Perfitt measures customers' foot shapes and sizes virtually, using user-provided images, and compares those measurements to various brands and designs available on the market. This reduces the risk of buying ill-fitting shoes and minimizes return costs for customers and businesses.
However, acquiring sufficient labeled data to train their models for accurate foot segmentation proved difficult, especially given the company's size and stage of product development. Moreover, Perfitt's raw and labeled visual data was dispersed across multiple systems, hindering management, visibility, and collaboration. To tackle these challenges and implement more data-centric workflows, Perfitt teamed up with Superb AI and adopted the platform and image labeling tools.
Perfitt experienced immediate improvements in data management, enabling the team to label and manage 90% of the company's data assets in one location. Collaboration also significantly improved, with in-house and external labelers, data PMs, and engineers now working together using a shared interface. Managers could easily monitor work throughput and quality, reducing data processing time and costs. The Perfitt team quickly identified issues such as miscommunication, leading to prompt enhancements in labeling guidelines and data outputs.
To further increase throughput, Perfitt adopted Superb AI's Custom Auto-Label feature and active learning workflows to further increase throughput, accelerating data segmentation by 1.5x, including QA time. Utilizing performance analysis and prediction metrics provided by the platform, instead of traditional trial and error, the team accurately determined the additional data needed for each object class to maximize model performance. This approach reduced labeling inefficiencies by eliminating the need to collect data for all classes.
Adopting Superb AI has enabled Perfitt to develop and implement a more efficient data processing strategy—a critical first step for any machine learning professional. Establishing a robust data pipeline allows Perfitt to continually refine its service model, adapting to external changes and customer needs. By leveraging the Superb AI platform, Perfitt maintains quality, reduces human labor, and saves time during the labeling stage, ensuring ongoing success in revolutionizing the online shoe shopping experience.