Creating SORDI: The Largest Synthetic Dataset for Industries

This chapter describes SORDI, the largest Synthetic Industrial Dataset for Object Detection for Industries, jointly developed by BMW Group and Idealworks. Created using Nvidia’s Omniverse, SORDI consists of more than 100 industrial assets in 35 scenarios and more than 1,000,000 photo-realistic rendered images that are annotated with accurate pixel-level bounding boxes. For evaluation purposes, multiple object detectors were trained with synthetic data to infer on real images captured inside a factory. Accuracy values higher than 80% were reported for most of the considered assets, highlighting the potential of the dataset and its practicality. Ch. 6 attempts to answer the following questions: How is SORDI built? How are the material designed? How are the images and scenes generated? How is the dataset used for object detection and recognition?