Providing Image Sequence Annotation for Autonomous Scene for Automotive Industry

  • Client Profile

    Industry: Automotive
    Location: Germany
    Size: 51-200

  • Company Bio

    The company is well-known in providing data preparation for world-leading OEMs and Tier 1 companies and requires support with data annotation, validation of pre-labeled data, and quality assurance. They requested us to help with the project of their client from the automotive industry.

Business Problem

Initially our partner came to us with the request to provide quality assurance (QA) services on the work done by another vendor as their client wasn’t satisfied with the data annotation quality level. As we demonstrated excellent results in QA, the partner later decided to pass over to us the entire project. It consisted of two phases: QA and annotation of large volumes of 2D image sequences with instance segmentation. The list of classes included 10 main positions with a large amount of attributes that dynamically change depending on the situation on the road, the landscape, car position, and other parameters.

Why Mindy Support

Mindy Support has in focus the success of its customers and partners and always strives to build long-lasting trust relationships with them. The partner had already a positive experience of working with us on several large projects for their customers, mainly from the automotive industry, and saw our ability to deliver excellent quality. It is also important to note that it was required to do the data annotation work in the partner’s tool and we already had experience with it, so it reduced the time for training the team and allowed us to perform the data labeling faster.

Solutions Provided

We allocated a QA team and trained our data annotation specialists to work with large amounts of 2D image sequences with the instance segmentation method. The main challenge we had to overcome was to define the right attribute for the object that was dynamically changing its position, occlusion level, and other parameters. While we were working with the partner’s tool, we also needed to build an additional extension to extract the necessary statistics from the partner’s tool and control the productivity metrics more efficiently. 

Results Delivered to the Client

  • The data annotation quality level increased from 92% to 98%
  • Saved partner’s and client’s time and resources