(Bloomberg) -- Merck KGaA, the German pharmaceutical and chemicals company, is teaming up with a London-based startup called V7 to create software that scientists can use to tag images for future use in training artificial intelligence software.
Many common forms of machine learning require large sets of labeled data to build expertise. But tagging objects in specialized scientific imagery or videos is time-consuming and often involves painstaking work by skilled professionals, such as radiologists outlining the boundaries of tumors in medical imagery or climate scientists identifying forests in satellite images.
“In health care alone, there are massive amounts of unstructured data generated every day in labs,” Michael Vande Vrede, head of digital products at Merck KGaA, said Tuesday at the conference on Neural Information Processing Systems (NeurIPS) in Montreal, where the two companies announced their collaboration.
V7 has created software called Graphotate that makes it easier and faster to annotate scientific imagers and videos. While experts are still required to provide these labels initially, with enough examples the software learns to automatically provide the annotation.
Alberto Rizzoli, the co-founder of V7, said that other object-recognition and tagging software, which has generally been designed for uses including autonomous vehicles and warehouse and factory management systems, wasn’t flexible enough for scientific uses. For instance, some only allow rectangular boxes to be used to designate the boundaries of an object and few had the ability to do something like circle florescence in a laboratory slide to incorporate both the identity of an object and a tag for its movement in the same annotation. Graphotate, he said, was designed to allow users to customize the tags and labels.
Existing software also wasn’t often designed to process the kinds of very large image files produced by specialized imaging equipment, Rizzoli added.
Graphotate, which will be released by March 2019, can run on a cloud-based data center or on local server in cases where companies are reluctant to have data leave their premises due to privacy and security concerns, he said.
Vande Vrede said Merck was helping support the project with funding, expert knowledge and by providing a large library of different kinds of scientific imagery. He said Merck itself would use the system once it was rolled out, but the company also wanted to help scientists create large open-source sets of images on which to train machine learning algorithms.
V7 was founded in 2015 in San Francisco as a company called Aipoly. The company built computer vision systems to help retailers pioneer cashierless checkout systems similar to the one Amazon.com Inc. is experimenting with. But Aipoly became best known for an app it created that allowed visually impaired people to use their smartphones to get audio descriptions of the objects seen through the mobile phone’s camera.
Merck, which is based in Darmstadt, Germany, was fascinated by the Aipoly app and started talking to the company about ways to collaborate on computer vision projects, Vande Vrede said.
The company, which has about a half dozen full-time employees, moved to London and changed its name to V7 in August.
To contact the reporter on this story: Jeremy Kahn in London at email@example.com
To contact the editors responsible for this story: Giles Turner at firstname.lastname@example.org, Andrew Pollack, Alistair Barr
For more articles like this, please visit us at bloomberg.com
©2018 Bloomberg L.P.