A self-supervised cyclic neural-analytic approach for novel view synthesis and 3D reconstruction
Dragoş Costea, Alina Marcu, Marius Leordeanu · 2025
Read A self-supervised cyclic neural-analytic approach for novel view synthesis and 3D reconstruction by Dragoş Costea, Alina Marcu, Marius Leordeanu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-03543.txt.
Description
Dragos Costea, Alina Marcu, and Marius Leordeanu present a method for rebuilding scenes in 3D from drone video, in this 2025 paper. New views far from the filmed path often come out poorly, which limits self-flying drones. Their pipeline loops between a neural renderer, which makes sharp images, and a classic geometry method, which keeps shapes honest.
The loop is self-supervised, so it needs no labeled training data. A transformer-based stage adapts the rendering to new viewpoints on the fly. Tests showed better image views and better mesh models than before, mainly in thinly filmed areas. The team frames the work as a step toward safer flight in complex outdoor settings.
Catalog record
| CREATOR | Dragoş Costea, Alina Marcu, Marius Leordeanu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-03543