GSH-Occ: Gradient-Shielded and Height-Aware BEV Occupancy Network
Bokai Ou, Tianhui Li, Zhigui Lin, Boao Wu, Pintong Chen, Zhajiacuo Zhou, Yating Liu, Jingyao Wang, Jinghua Guo, Lei He · 2026
Read GSH-Occ: Gradient-Shielded and Height-Aware BEV Occupancy Network by Bokai Ou, Tianhui Li, Zhigui Lin, Boao Wu, Pintong Chen, Zhajiacuo Zhou, Yating Liu, Jingyao Wang, Jinghua Guo, Lei He (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/pmc-13166031.txt.
Description
Bokai Ou and nine co-authors build a camera network that maps the space around a car in three dimensions, fixing two known flaws of the bird's-eye approach. Occupancy prediction tells a self-driving car which voxels are filled, and most systems lift a top-down view into 3D. That lift usually ignores height, and bolting new modules onto pretrained encoders destabilizes training. Their 2026 network, GSH-Occ, answers with a gradient-shielded attention block that starts as a pass-through, so new parts do not wreck old training, plus a height-aware lifting that keeps elevation detail.
Both fixes are surgical: change the grafting, change the lift, keep the pipeline. Occupancy maps are becoming the common language between perception and planning. The authors write for the autonomous driving vision community. Their design respects a practical truth: most teams improve pretrained systems rather than starting over.
Catalog record
| CREATOR | Bokai Ou, Tianhui Li, Zhigui Lin, Boao Wu, Pintong Chen, Zhajiacuo Zhou, Yating Liu, Jingyao Wang, Jinghua Guo, Lei He |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |