FRONTISPIECE . Enabling neural radiance fields for large‐scale aerial images – From a geometric reconstruction perspective
Year: 2024
Type: paratext
Abstract: The paper in this issue proposes a multi-camera tiling approach to perform 3D reconstruction using a neural radiance field (MCT-NeRF), which delivers superior reconstruction results for complex objects. This figure visually compares the MCT-NeRF method with other traditional reconstruction methods. The rows correspond to complex and thin objects with widths of only a few pixels in the image: (1) a stone pillar (7–11 pixels in width), (2) a light bulb (4–5 pixels lateral resolution), (3–4) a thin-frame structured arched door (each frame bar takes only 2–3 pixels in width), and (5) a group of thin pillars (2–3 pixels in width) with heavy occlusions. Our MCT-NeRF consistently outperforms the other methods in reconstructing these fine structures. For more details see Xu and Qin (2024) in this issue. The code is available on GitHub.
Source: The Photogrammetric Record
DOI: 10.1111/phor.12527
Language: English
Subfield: Geology
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4405184221