Artificial intelligence startup Hasty GmbH said today it has nabbed $3.7 million in seed funding as it looks to speed up the development of its next-generation annotation tool for computer vision.

Shasta Ventures led the seed funding round, together with participation from coparion and iRobot Ventures.

Hasty has built an annotation tool that’s used to train vision AI models that are used by computers to understand what they’re seeing when presented with an image or video. In order for a computer to know what it’s seeing, the content of the image or video needs to be labeled, so it can “see” different objects or human faces and read text, for example.

But data labeling is a laborious task, and it’s hindered by the fact that most vision AI models don’t train and update themselves during this process. Indeed, Hasty says that around 80% of a data scientist’s time is spent on labeling the data they use to train their neural networks. Because of this delay between labeling and model training, machine learning engineers often have to wait for months before they can see if their annotation strategy is actually working.

Hasty says its next-gen annotation tool adds more haste to vision AI training because it can train models while the data is being labeled. Moreover, the tool becomes faster at labeling objects the more it’s used, because the model helps create an automation cycle by informing the labeling process once it has started learning.

Hasty said its AI Assistants are able to predict what labels should be after just a few images have been annotated, enabling data scientists who are building their data sets to work much faster. The tool also comes with an Error Finder & Manual Review bot that helps detect annotation errors from misclassifications, artifacts, missed objects and poor segmentation.

Meanwhile, the Rapid Feedback Loop bot provides short feedback cycles that enable data scientists to iterate faster and validate and adapt their data labeling approach as necessary. That means they can test various approaches to be sure they’re creating the best data to train each model.

“We put our customers ahead of the game and significantly increase their ability to bring more successful vision AI projects to market faster,” said Hasty co-founder and Chief Executive Tristan Rouillard.

Hasty believes there will be big demand for any tool that can speed up the vision AI training process, because the discipline has enormous potential in so many industries. Some examples include helping physicians to diagnose diseases by looking at X-rays or scans more quickly, or tracking packages through supply chains.

“Vision AI has the power to unlock the future of automation and accelerate the creation of new products and services in a way not seen since the Web Revolution,” said Shasta Ventures Partner Isaac Roth. “Hasty’s team of engineers understands this and has delivered a best-in-class solution that gets faster and smarter the more it is used.”

Image: Hasty