Pixel-Based Foreground & Background Image Segmentation

The Problem

Our client, a multinational technology company, approached DataForce looking for a partner to help scale its object lifting and detection image segmentation. The company needed to generate high quality training data to support one of its upcoming augmented reality (AR) based product releases. Our client required a highly scalable and cost-effective lab-based segmentation solution that would deliver exceptional training data quality paired with a partner that could deliver while working on both diverse and complex image subjects, without putting their release schedule at risk.

The Solution

Using our global footprint of offices and secure labs in over 120 cities and 46 countries, DataForce first identified a suitable offshore location to drive cost savings, and quickly proceeded to train and deploy a team of over 100 annotators.

Working closely with the client and applying our deep industry knowledge and proprietary DataForce technology, we were able to advise and select the best image segmentation method. By leveraging some of the advanced annotation features built into our DataForce platform—such as intelligent scissors, grab cut, and super pixel, in addition to highly customizable polygon annotation—we were able to quickly identify the most suitable annotation method that best aligned to the diverse and complex image subjects.

We took a collaborative approach with the client during the configuration of the project, allowing us to train the team in real time and significantly exceed our client’s quality expectations, delivering several hundred thousand accurately annotated images over the course of two months. The images were provided to the client to fully train their machine learning model, significantly improving the user experience ahead of launch, and enabling the timely release of their AR product.