VidiNet, the cloud-based media services platform from Arvato Systems’ Vidispine portfolio now comes with even more features. The integration of Deep VA’s AI software into the system allows users to recognise information in image or video files and create their own AI models. In addition, the integration with VidiNet Cognitive Services also allows for the creation and quality assurance of training data within a familiar media asset management environment.
VidiNet provides a broad range of media services in a pre-integrated environment. It allows users to create media workflows of any kind – with unlimited scaling options. VidiCore is the media management backend and object repository and forms the basis of Vidispine’s media asset management solutions. Vidispine itself represents a complete portfolio to create, produce, prepare, manage and monetise media content. This includes proprietary applications and tightly integrated applications from our partners.
Thanks to this enhancement, VidiNet and VidiCore users now also have access to DeepVA’s AI software. The software is delivered through VidiNet Cognitive Services, an interface that provides VidiCore customers with direct access to software offerings around cognitive services provided by partners. The integration of DeepVA tremendously simplifies media workflows across the content ecosystem and provides users with an easy access to AI within their familiar MAM environment.
VidiNet users profit from this attractive addition to the Vidispine application portfolio as the integration greatly facilitates the step toward AI, which will be indispensable in the near future, without adding complexity – no expert knowledge is needed. DeepVA’s AI enriches existing image and video material with metadata which greatly simplifies and accelerates the documentation and search of media assets. One big advantage: so-called „custom AI models“. With DeepVA, AI models can be trained according to individual requirements. It also allows to find information in media assets, a task that lets “pre-trained models“ by large providers often reach their limits.
Face and label extraction can be used to automatically extract training data from videos and livestreams by linking names inserted in the image with the corresponding faces and storing them in a dataset. That way, individual training data is generated automatically within seconds, which then forms the basis for individual AI models and thus the recognition of media-specific information in image and video material. Due to the so-called ‘one-shot learning’, it usually only requires one image to train an AI model, which is a huge time saver when building training data.
DeepVA is already used by large broadcasters, DAM/MAM providers, streaming services and city archives, which produce significant amounts of videos and images on a daily basis. DeepVA covers a large part of the overall AI application areas such as visual concept recognition, face recognition, landmark recognition, brand/logo recognition as well as text recognition.
Within VidiNet, it is possible to access the Face Recognition, Custom Faces (individual creation of AI models of faces), Dataset Creation (automated creation of training data) as well as the Face Indexing models.
The service is billed on a pay-as-you-use basis, i.e. users are granted a volume license, which is billed per analysed image or analysed video minute as well as by module.