skip to Main Content
What We Do

Through Dense Pixel Segmentation, Moving Bounding Boxes and Facial Key Points, our team helps build the training data that powers the world’s most advanced image and video algorithms.

How We Do It

Extensive experience: over 7 million images segmented and annotated.  Custom workforce training. Dedicated project management and tools team. Edge case insights. 100% quality control.

Who We Do it For

Our customers use Computer Vision in diverse fields including autonomous vehicles, aerial imagery analysis, medical imagery, sports and mixed reality.

Semantic Segmentation
Street Scenes for Driverless Cars

The client is a leading global automobile manufacturer and a major contender in the autonomous vehicles segment. They need tremendous volumes of data, segmented pixel by pixel, into meaningful classes of precise objects. This is used to train the machine learning algorithm for their driverless cars and help the vehicles understand their environment completely for a smooth and safe drive.

iMerit employs a team of visual data experts that performs semantic segmentation on thousands of street images everyday. They label these images pixel by pixel into pre-determined classes of objects, ultimately dividing the image into semantically meaningful parts. This is done with the utmost precision and reviewed by quality experts in parallel to ensure >95% accuracy.

Our team has segmented over 100,000 images so far and continues to work with a similar volume now as the customer iterates and expands their effort. These image datasets help train the machine learning algorithm not just “see” but also interpret its environment.

World Series-Winning Machine Vision

KinaTrax has a markerless motion capture system used by Major League Baseball (MLB) teams to measure their players’ performance, fatigue, and risk of injury. KinaTrax captures in-game footage of players  from twelve different 4K high frame rate cameras. KinaTrax  requires precise joints registered in 3D in order to run machine vision and biometric algorithms to analyze the movements of MLB players. MLB teams can then use these analyses in order to make decisions about player health, performance, and safety.

iMerit teams are involved in the KinaTrax process from end-to-end. We first process in-game video footage of players immediately after each game. We select suitable candidate frames for annotation and then annotate precise joint positions from multiple angles. These images are fed back into KinaTrax processes for analysis and distribution to MLB teams within two days.

iMerit teams have created accurate 3D models for over 300 MLB pitchers. These models are used as foundations for ongoing in-game analysis by KinaTrax and were used by the 2016 Chicago Cubs in their historic World Series win. KinaTrax is expanding its work to other players in the game.

Back To Top