Geo-Code: A Code Framework for Reverse Code Generation from Geometric Images Based on Two-Stage Multi-Agent Evolution
Zhenyu Wu, yanxi long, Jian Li, Hua Huang · 2026
Read Geo-Code: A Code Framework for Reverse Code Generation from Geometric Images Based on Two-Stage Multi-Agent Evolution by Zhenyu Wu, yanxi long, Jian Li, Hua Huang (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-2602-07749.txt.
Description
Turning a geometry diagram back into code could teach models to reason about shapes, but existing methods lose constraints and distort structure. Zhenyu Wu and three co-authors propose Geo-Code, a multi-agent framework that reconstructs programs from geometric images in two stages.
First, visual operators and large models work together to pin down pixel coordinates and visual attributes. Then a loop of synthesis, rendering, and validation uses visual feedback to correct the code itself. Experiments show the approach captures geometric detail more faithfully than prior inverse-graphics methods. The work matters for multimodal AI, where a program that redraws a diagram is strong evidence the model truly understood it.
Catalog record
| CREATOR | Zhenyu Wu, yanxi long, Jian Li, Hua Huang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07749