Largest Open-Source Synthetic Dataset for AI Applications

May, 2022

In collaboration with well-known national and international players from the automotive and technology sectors, idealworks developed the dataset Synthetic Object Recognition Dataset for Industries (SORDI). The dataset consists of more than 800,000 photorealistic images of production resources in 80 classes and includes objects of particular relevance in the core technologies of automotive engineering and logistics.

Together with its long-term partner NVIDIA Corporation, Microsoft Corporation, and the BMW Group’s BMW TechOffice MUNICH, idealworks is about to release SORDI, the world’s largest reference dataset for artificial intelligence in manufacturing. Specifically designed for industrial environments, such as manufacturing and production, it significantly accelerates neural network training for device recognition in the factory environment and makes them even more efficient.

The BMW Group has already been deploying AI in the production system since 2019: “Artificial intelligence is used productively in quality assurance in various applications in our plants. The new, synthetic data set SORDI allows much faster training of AI models and thus serves to significantly increase the cost efficiency of AI in production,” says Michele Melchiorre, SVP BMW Group Production System, Technical Planning, Tool Shop, Plant Construction.

The BMW TechOffice MUNICH’s rendering pipeline ensures the automatic creation of any number of synthesized images in photo-realistic HD quality including labels, which in turn allows for the realization of AI models with a high degree of robustness. In addition, integrated digital labels enable fundamental machine vision tasks for relevant areas of production, such as classification, object recognition, or segmentation.

The robotics simulation environment, the digital twin of the production system, and the AI training environment are consolidated in NVIDIA Omniverse for a smooth creation of artificial intelligence based on synthesized data without much manual effort.

Synthesizing AI based on SORDI can be applied in all kinds of technologies: “At idealworks, we use the algorithms to develop, simulate, and secure our autonomous mobile robot iw.hub. We deliberately included resources such as dollies or shelves, which play an important role in the further development of our AMR,” explains idealworks’ CTO Jimmy Nassif.

Marc Kamradt, Head of BMW TechOffice MUNICH, adds, “SORDI enables us to train a production-specific AI backbone, providing us with an individual basis for AI applications specifically in production.”

The dataset facilitates the work of production staff by providing them with mature AI systems to back them up in time for the start of production, and also supports IT experts in developing and customizing AI solutions. SORDI will soon be available to software developers open-source via GitHub at the following link: github.com/bmw-innovationlab. Visit sordi.ai to follow along and keep up to date.