Generative Adversarial Network (GAN) for Synthetic Data Generation

To advance state-of-the-art (SOTA) landscape and object synthetic data generation through the development of a generative adversarial network (GAN), to produce infinite training data compatible with bi-directional mapping, dynamic multi-object tracking, and baseline obfuscation use cases. When commercialized and applied to scale, this technology will enable a variety of industries, from agriculture, transportation, retail, or humanitarian aid, to train machine learning (ML) models on realistic geospatial landscapes and simulations.  VISIMO will utilize a unique combination of techniques previously proven individually to create a singular image-generation model. Various methods of image segmentation, annotation, and image-inpainting will be tested to identify the ideal combination of techniques suitable to form the training data for the GAN’s discriminator.