Unconstrained Large-scale 3D Reconstruction and Rendering across Altitudes
This addresses the need for accessible 3D reconstruction tools for first responders in disaster relief or law enforcement, though it is incremental as it focuses on dataset creation rather than a new method.
The paper tackled the problem of producing photorealistic 3D models from limited and heterogeneous images across altitudes by creating the first public benchmark dataset for 3D reconstruction and novel view synthesis, demonstrating baseline performance with state-of-practice methods and identifying research challenges.
Production of photorealistic, navigable 3D site models requires a large volume of carefully collected images that are often unavailable to first responders for disaster relief or law enforcement. Real-world challenges include limited numbers of images, heterogeneous unposed cameras, inconsistent lighting, and extreme viewpoint differences for images collected from varying altitudes. To promote research aimed at addressing these challenges, we have developed the first public benchmark dataset for 3D reconstruction and novel view synthesis based on multiple calibrated ground-level, security-level, and airborne cameras. We present datasets that pose real-world challenges, independently evaluate calibration of unposed cameras and quality of novel rendered views, demonstrate baseline performance using recent state-of-practice methods, and identify challenges for further research.