GA-Drive: Geometry-Appearance Decoupled Modeling for Free-viewpoint Driving Scene Generation
Hao Zhang, Lue Fan, Qitai Wang, Wenbo Li, Zehuan Wu, Lewei Lu, Zhaoxiang Zhang, Hongsheng Li, [0.5em] MMLab, CUHK CASIA · 2026
Read GA-Drive: Geometry-Appearance Decoupled Modeling for Free-viewpoint Driving Scene Generation by Hao Zhang, Lue Fan, Qitai Wang, Wenbo Li, Zehuan Wu, Lewei Lu, Zhaoxiang Zhang, Hongsheng Li, [0.5em] MMLab, CUHK CASIA (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-2602-20673.txt.
Description
A free-viewpoint, tunable, high-quality driving sim matters for training and testing end-to-end autonomous driving systems. Hao Zhang and seven co-authors present GA-Drive, a simulation system that makes camera views along user-chosen novel paths through shape-look split and diffusion-based making. Given images from a filmed path plus scene shape, GA-Drive builds novel pseudo-views from shape information.
Those pseudo-views are then turned into lifelike views by a trained video model. Splitting shape from looks enables look editing through state-of-the-art video-to-video editing techniques while keeping base shape, so edits stay steady across first and novel paths. Broad tests show GA-Drive clearly beats current methods on NTA-IoU, NTL-IoU, and FID scores.
Catalog record
| CREATOR | Hao Zhang, Lue Fan, Qitai Wang, Wenbo Li, Zehuan Wu, Lewei Lu, Zhaoxiang Zhang, Hongsheng Li, [0.5em] MMLab, CUHK CASIA |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-20673