From Orbit to Ground: Generative City Photogrammetry from Extreme Off-Nadir Satellite Images

The project addresses a critical bottleneck in geospatial intelligence: the '90-degree gap' where satellite images lack the vertical facade data necessary for immersive 3D city models. While standard 3DGS and NeRF fail due to extreme foreshortening, this approach uses generative priors to hallucinate missing textures. The defensibility is currently low (score 4) because it is a reference implementation of a paper; the primary moat in this space is access to high-resolution satellite constellations and ground-truth oblique aerial data (e.g., Maxar, Vexcel), not just the reconstruction algorithm. The high fork-to-star ratio (13 forks, 0 stars) suggests immediate interest from the academic and defense research community despite the repo's age. Platform risk is high because companies like Google (Google Earth) and Microsoft (Bing Maps) already possess the massive datasets and compute required to bake this into their existing 3D tile pipelines. For a startup, the value lies in specialized GEOINT applications (e.g., drone simulation, tactical urban planning) where standard satellite imagery must be rapidly converted into navigable 3D environments.