The project has indirectly received funding from the European Commission’s Horizon 2020 Framework Programme through the STADIEM project (Grant Agreement 957321)
CUSTOMER SUCCESS STORY / VIDISPINE
Artificial intelligence in the media supply chain
Enhancement of a media management system with automatic image and video recognition as well as an integrated feature that allows creating individual AI models in no time.
With the help of DeepVA, the user can intuitively and easily create and manage training data directly in the media asset management system (MAM).
This greatly increases identification performancein the field of face recognition.
Arvato Systems is an internationally active IT specialist and multi-cloud service provider whose mission is to support various companies in their digital transformation. Fast and secure IT systems provide Arvato’s customers easy access to the cloud and ensure optimal connected applications and business processes with the help of IoT, blockchain or artificial intelligence.
With VidiNet, Arvato Systems offers a cloud-based media service platform of the Arvato Systems brand Vidispine, which constitutes a highly efficient basis for the entire content chain with its numerous applications. It offers users a range of services and apps in a pre-integrated environment as SaaS solutions. Thus, the platform not only supports various media workflows, but also allows any scaling options for professional use.
The cooperation of our AI software with the Arvato Systems Vidispine team turned out to be a complete success. From the beginning, both we and the team behind Arvato Systems were fascinated by the idea of automating media workflows with the most advanced tools from the IT toolbox, namely AI in the field of computer vision. The user of the media asset management system should be able to monitor and control the recognition of content from images and video, the creation of their own AI models and the quality of their training data themselves.
The implementation of prefabricated AI models in the field of computer vision rarely results in the hoped-for added value in practice. While integration into a MAM can be straightforward and fast, recognition performance is severely limited. For example, most organizations additionally need to recognize people in images and videos that are not part of an existing AI model. To achieve this, models must be constantly expanded, updated and adapted. The necessary tools for this must be made transparently available to the user directly in the MAM,
because developing your own solutions to adjust AI models to your own company’s needs is usually expensive and requires expert knowledge in the field of machine learning.
DeepVA, in contrast to other business organizations in the field of computer vision, offers the possibility to automatically create own AI models, resulting in more individual content that can be recognized from media assets compared to “pre-trained models”.
The cooperation between DeepVA and VidiNet allows for building individual AI models in the MAM system without any prior technical knowledge. Due to the growing amount of available training data, which is considered the gold of the data-driven age for good reason, the performance of person recognition is increasing rapidly. This results in a more detailed and higher quality tagging of images and videos and as a result in an improved searchability of all content. In addition, face indexing (assigning a unique ID to unrecognized persons) enables further analyses and reverse searches. This optimizes workflows in the MAM and greatly saves time and costs. Employees must no longer spend their time on monotonous and repetitive manual tagging of media content, but instead create an intelligent and constantly improving visual data management system with just a few clicks.
DeepVA, VidiNet, and the accompanying VidiNet Cognitive Services, create a highly integrated MAM ecosystem that allows users to build their own AI models in a simple and intuitive way. With the integration of DeepVA in VidiNet, the required training data can be efficiently created and managed. Sample data can be provided manually or cut and labeled directly from videos using a tool. Face Dataset Creation even allows to automatically generate appropriate training data, by extracting and saving sample images from interview scenes. In addition to the management of training data, the data can also be automatically checked for usability for custom AI models.
FASTER DATA ACQUISITION
NEED OF AI EXPERTS