A medical artificial intelligence (AI) company that provides an image database for research and evidence-based care, needed a way to label thousands of images quickly and accurately to further its product offering.
Managing in-house staff, even on a contract basis, was time-consuming and expensive. But the image files are big and the task complex, so outsourcing the task seemed daunting as well. Within months of choosing CloudFactory’s managed workforce option, the company has completed 12 additional imagery databases to improve their product and is sold on the managed workforce process.
"The space that I’m in is going through an evolution. If you don’t have AI capabilities, you’ll be left behind. But I can’t focus on the product if I’m swamped doing people management. CloudFactory takes that burden off of us."
The medical AI company needed the labeling work done by a consistent group of individuals who could log-on remotely to the company’s own annotation tool. This is because the labeling tool is custom-designed with machine learning components built-in.
The work is critical to the company’s effort to stay ahead of the curve in providing AI-based image databases that enhance medical professionals understanding of health issues. One of the company’s goals is to provide predictive advice based on tagging and analyzing images throughout the patient’s lifespan. The goal is to enhance preventative care.
The company’s owner researched multiple companies looking for image tagging help.
“I was worried about crowdsourcing because I didn’t think the quality would be there. And some of the other companies I looked at had preconceived ways of working with you,” he explains, noting some were not interested in having staff work directly on his platform. In addition, he needed workers used to dealing with images because each batch is unique in terms of what is being tagged.
The company chose CloudFactory for its flexibility and its experience working with other computer vision companies. CloudFactory workers are trained in a wide array of use cases from bounding boxes and semantic segmentation to 3D point cloud and sensor fusion systems. In addition, CloudFactory has a set of best practices developed from annotating millions of images and videos over the last ten years.
CloudFactory used its “train the trainer” approach where the client can work closely with a staff member who will be working directly with the managed workforce. Initially, the owner was a little bit concerned that data labelers without a medical background would be successful, “but I thought it was possible.’’
“I was pleasantly surprised. I did write some documents and record some screen videos. That’s all,’’ the company owner says adding, “They’ve done a good job. But if something isn’t quite right, I give feedback and they get to the higher bar for the next set of data. They’ve also given us valuable feedback on our tool.’’
He thinks it works well because CloudFactory provides a U.S.-based manager and a team lead where the workers are based (Nairobi). “I like having that arrangement.”
The altruistic aspect of CloudFactory’s approach also wooed the owner. “You’re on a bit of a mission to help people around the world that are good technologically but don’t have the opportunity. That was a big factor in my decision because we’re also trying to do the right thing.’’
CloudFactory was up to full capacity on the company’s project within just two weeks and were self-sufficient after three months, giving the owner more time to focus on the go-to-market side of his business.
“The space that I’m in is going through an evolution. If you don’t have AI capabilities, you’ll be left behind. But I can’t focus on the product if I’m swamped doing people management. CloudFactory takes that burden off of us.’’