Gen Studio (generator of images)

2019: Announcement of technology

At the beginning of March, 2019 Microsoft announced the project on generation of images on the basis of works of art. The company developed technology together with Metropolitan Museum and the Massachusetts Institute of Technology (MIT)

Using public API interfaces, images and a key word, specialists submitted the program under the name Gen Studio allowing to combine different objects from Metropolitan Museum collection. Service on the basis of generative and competitive network (GAN) is also combined with visual search that allows users to study a collection and to open new areas of art space.

Microsoft provided the generator of images on the basis of works of art

GAN capable to find the closest compliance between the set image and this work of art in Metropolitan Museum is the cornerstone and to combine them in uniform work. Besides, users can independently mix different works – developers hope that it will help people to understand better the visual structure which is the cornerstone of a museum collection and also to create and combine the artworks based on different styles, materials and forms.

For creation of this service developers used microservice architecture of deep neuronets, services Azure and storage of BLOB objects. Visual Studio Code and Azure Kubernetes Service allow to create new images in real time and are responsible for interactive appearance of the website.

Azure Kubernetes Service significantly simplifies production, allowing to deploy the service for only several days. That users could study Metropolitan Museum collection, developers loaded all available images into a cluster of Azure Databricks and used machine learning of Microsoft for Apache Spark to supply these images with descriptions. Using ResNet50 quick search on visual similarity was created, and data from a cluster were loaded into Azure retrieval service.^[1]

Robotics

Notes