Research project: KIMO

Digital innovation and technology

The aim of the project is to provide SMEs with low-threshold access to cutting-edge technologies from the field of deep learning (artificial intelligence) for individual pattern and object recognition in images. Previous methods such as the YOLOV4 require a very large image data set (over 200,000 images) for the learning process to perform well. However, SMEs can often only provide a small dataset of annotated recordings for training. This data set consists mostly of 30 to 50 images. Due to this small data set, classical training methods often cannot be applied. The use of transfer learning methods, which have already been successfully applied in the field of person recognition, is intended to reduce the number of images required to a minimum. In addition, the training process, which today must be accompanied manually by a machine learning engineer, is to be automated as far as possible.

Scope

Automatic pattern and object recognition can accelerate very time-consuming manual processing steps in the management, distribution and provision of images and videos (especially for large media volumes) in the teamnext | Media Hub. These could be, for example, the automatic recognition of sponsor logos in professional sports, of products in advertising materials, of regional tourist attractions or of components for spare parts procurement. In addition to direct marketing as a technology component in the teamnext | Media Hub, it is planned to make the service for AI-supported pattern and object recognition usable for third-party systems via an API interface. This would make the new process available to other technology SMEs for use in apps and web applications.

Project participants

The teamnext GmbH & Co KG was founded in 2010 and is a software-as-a-service provider in the field of cloud-based digital asset management (DAM). With the teamnext | Media Hub, an AI-powered DAM platform is offered that helps organizations better structure digital assets (images, videos, and graphic files), automate delivery, and simplify organization-wide distribution. A particular focus of teamnext is the development of its own AI services for the efficient management of image and media files, which can be highly customized and optimized for specific use cases. In 2020, a proprietary AI strategy was implemented with no dependencies on or limitations from third-party solutions. In recent years, various individual problems in the field of machine vision have been successfully solved on behalf of customers. Customers include the German Association of the Automotive Industry (VDA), the Free University of Berlin, the German Olympic Sports Confederation (DOSB), the University of Duisburg-Essen, the German Handball League, the International Motor Show (IAA) as well as numerous SME market leaders, public bodies, associations and listed companies from the GSA region.