Synthetic content can be artificially produced, manipulated, or modified using artificial intelligence (AI) to create a wide range of synthesized data from text, prose, music, or images to videos. When used for a malign purpose to mislead or deceive, this synthetic content has come to be known as “deepfakes”. But beyond the potentially sinister, headline-grabbing generation of fake videos of world leaders, deepfake technology can be applied to create content and data valuable to the construction of synthetic environments.
Data describing the physical world is now very abundant and provides a rich data source to support synthetic environment development. Still, there are many scenarios where this data may not be available or suitable. In these situations, artificially created data can provide data where none exists, augment existing data, improve data quality to provide accurate representations of real features, data and places, or even create entirely fictitious data and environments.
Generative artificial intelligence is one of the most promising advances in AI in the past decade. Generative models produce synthetic content by learning to mimic a data distribution and generate new, similar, credible content. For example, they can create entirely artificial satellite imagery, increase image resolution or remove artefacts with inpainting. They can also generate or manipulate other data types, such as land cover, point clouds, maps, 3D models, or even the design and style of the entire environment, all of which will appear authentic but are, in fact, entirely artificial
This paper will explore the current state and capabilities of generative models. It will identify those model types, such as Generative Adversarial Networks, that are most suited to generating synthesized content for synthetic environments. Finally, it will illustrate how these models can be applied to specific types of content and use cases and evaluate the currently achievable results.
Keywords
AI,CONTENT GENERATION,DEEP LEARNING,GEOSPATIAL DATA ,MACHINE LEARNING,SYNTHETIC ENVIRONMENT
Additional Keywords