CoLoGen: Progressive Learning of Concept–Localization Duality for Unified Image Generation

Yuxin Song, Yu Lu, Haoyuan Sun, Huanjin Yao, Fanglong Liu, Yifan Sun, Haocheng Feng, Hang Zhou, Jingdong Wang · 2026

Read CoLoGen: Progressive Learning of Concept–Localization Duality for Unified Image Generation by Yuxin Song, Yu Lu, Haoyuan Sun, Huanjin Yao, Fanglong Liu, Yifan Sun, Haocheng Feng, Hang Zhou, Jingdong Wang (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-2602-22150.txt.

Description

Unified image generators must juggle tasks that need different inner representations: some need concepts, others need precise locations. YuXin Song and eight co-authors argue that forcing one shared representation creates a concept-localization conflict. Their framework, CoLoGen, learns the two sides progressively instead, reconciling them for unified diffusion-based generation.

A staged curriculum first builds core concept and location skills, then adapts them to varied visual conditions, and finally refines their teamwork for complex instruction-driven tasks. A routing module sends features to specialized experts and blends their outputs steadily across stages. Experiments on editing, controllable generation, and customization show the staged design paying off.

Catalog record

CREATORYuxin Song, Yu Lu, Haoyuan Sun, Huanjin Yao, Fanglong Liu, Yifan Sun, Haocheng Feng, Hang Zhou, Jingdong Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-22150

Browse the full library · About the Tower of Babel