Wid3R: Wide Field-of-View 3D Reconstruction via Camera Model Conditioning
Dongki Jung, Jaehoon Choi, Adil Qureshi, Somi Jeong, Dinesh Manocha, Suyong Yeon · 2026
Read Wid3R: Wide Field-of-View 3D Reconstruction via Camera Model Conditioning by Dongki Jung, Jaehoon Choi, Adil Qureshi, Somi Jeong, Dinesh Manocha, Suyong Yeon (2026) 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-2602-05321.txt.
Description
Wide-angle and 360-degree cameras break the assumptions of most 3D reconstruction software, which expects tidy pinhole images. Dongki Jung and 5 co-authors present Wid3R in this 2026 computer vision paper, a feed-forward network that reconstructs multi-view geometry directly from wide field-of-view imagery. No explicit calibration or undistortion step is needed before the network sees the pictures.
The design represents viewing rays with spherical harmonics and adds a camera model token, so the network knows which distortion family it is dealing with and can correct for it internally. The authors describe Wid3R as the first multi-frame feed-forward method supporting full 360 imagery. Training across diverse camera types also improved generalization and eased data scarcity for 360 scenes. On a fisheye benchmark the accuracy score rose by up to about 34 points, and on a 360 indoor dataset by up to about 77 points, over the comparison methods.
Catalog record
| CREATOR | Dongki Jung, Jaehoon Choi, Adil Qureshi, Somi Jeong, Dinesh Manocha, Suyong Yeon |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-05321