TinyDoNetsk (GTA7)

Reconstructing and simulating an active warzone (Donetsk, Ukraine) in a game engine using satellite and topology data for ML training.

In TinyDoNotesk, also known as GTA7, we reconstruct and simulate 20 square kilometers of an active warzone (Donetsk, Ukraine) in the Unity game engine using satellite and topology data. We then use this simulated world to algorithmically generate and label targets (enemy vehicles) so that we can assemble a synthetic dataset to train ML equipped drones. Our final synthetic dataset consists of over 2,000 samples. Finally, we train an object detection model on our synthetic dataset and use TinyML techniques to shrink the model so that it can fit on a drone's edge compute.