Large sets of synthetic image data are becoming the backbone of training an artificial intelligence in the domain of computer vision. For a long time, the question has not been whether to use them but how to optimally create them. Using our own thirty years of practical experience in synthesizing graphics, we researched the best practices of generating synthetic image data, systematized the methods and techniques used, and brought specific recommendations to the creators of future synthsets. This exhaustive research by our director of research and development, under the mentorship of Associate Professor Marina Ivašić-Kos, was published in the prestigious journal Artificial Intelligence Review
- we recommend it to everyone who trains computers how to look and see!