Perceive-then-Plan: Layout-as-Policy for Monocular 3D Scene Layout Estimation
Junwei Zhou, Yu-Wing Tai · 2026
Read Perceive-then-Plan: Layout-as-Policy for Monocular 3D Scene Layout Estimation by Junwei Zhou, Yu-Wing Tai (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-2605-25326.txt.
Description
Junwei Zhou and Yu-Wing Tai turn single-image 3D layout estimation into a perceive-then-plan task. A Perceiver first grounds objects in 3D, then a Planner refines the scene for physical plausibility while staying faithful to the image. Their Layout-as-Policy approach treats layouts as states and corrections as actions like shifting, rotating, and rescaling. Training mixes trajectory starts with preference tuning, so the model learns fixes without hand-built rewards. Experiments show layouts that hold together physically and match observations, with support for editing and manipulation.
Catalog record
| CREATOR | Junwei Zhou, Yu-Wing Tai |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54644 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.25326