Startup Accelerates Object Labeling of Image and Video Files for Machine Learning

The startup understand.ai improves labeling, also called annotation, of single-image elements such as a tree, a pedestrian, or a road sign to accelerate machine learning for autonomous vehicles. According to the company’s founders, objects on the images used for machine learning are currently being labeled manually by human staff.

Using processed images, algorithms learn to recognize the real environment for autonomous driving. Courtesy of understand.ai.

The algorithms used to train autonomous vehicles draw from a large number of image and video recordings; and the labeling of these image and video files must be accurate down to the pixel level.

“We at understand.ai use artificial intelligence to make labeling up to 10× quicker and more precise,” said co-founder Philip Kessler, a graduate of Karlsruhe Institute of Technology.

The company also offers simulations based on real data for situations such as accidents, for which training images cannot be supplied. Although the startup is currently focused on labeling for autonomous driving, it plans to process image data for training algorithms to detect tumors and to evaluate aerial photos in the future.

Understand.ai is headquartered in Karlsruhe, Germany, with offices in Berlin and San Francisco.