FocalOrder: Focal Preference Optimization for Reading Order Detection
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911369331212288 |
|---|---|
| author | Liu, Fuyuan Yu, Dianyu Ren, He Liu, Nayu Kang, Xiaomian Qiu, Delai Zhang, Fa Zhen, Genpeng Liu, Shengping Liang, Jiaen Huang, Wei Wang, Yining Zhu, Junnan |
| author_facet | Liu, Fuyuan Yu, Dianyu Ren, He Liu, Nayu Kang, Xiaomian Qiu, Delai Zhang, Fa Zhen, Genpeng Liu, Shengping Liang, Jiaen Huang, Wei Wang, Yining Zhu, Junnan |
| contents | Reading order detection is the foundation of document understanding. Most existing methods rely on uniform supervision, implicitly assuming a constant difficulty distribution across layout regions. In this work, we challenge this assumption by revealing a critical flaw: \textbf{Positional Disparity}, a phenomenon where models demonstrate mastery over the deterministic start and end regions but suffer a performance collapse in the complex intermediate sections. This degradation arises because standard training allows the massive volume of easy patterns to drown out the learning signals from difficult layouts. To address this, we propose \textbf{FocalOrder}, a framework driven by \textbf{Focal Preference Optimization (FPO)}. Specifically, FocalOrder employs adaptive difficulty discovery with exponential moving average mechanism to dynamically pinpoint hard-to-learn transitions, while introducing a difficulty-calibrated pairwise ranking objective to enforce global logical consistency. Extensive experiments demonstrate that FocalOrder establishes new state-of-the-art results on OmniDocBench v1.0 and Comp-HRDoc. Our compact model not only outperforms competitive specialized baselines but also significantly surpasses large-scale general VLMs. These results demonstrate that aligning the optimization with intrinsic structural ambiguity of documents is critical for mastering complex document structures. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_07483 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | FocalOrder: Focal Preference Optimization for Reading Order Detection Liu, Fuyuan Yu, Dianyu Ren, He Liu, Nayu Kang, Xiaomian Qiu, Delai Zhang, Fa Zhen, Genpeng Liu, Shengping Liang, Jiaen Huang, Wei Wang, Yining Zhu, Junnan Computer Vision and Pattern Recognition Reading order detection is the foundation of document understanding. Most existing methods rely on uniform supervision, implicitly assuming a constant difficulty distribution across layout regions. In this work, we challenge this assumption by revealing a critical flaw: \textbf{Positional Disparity}, a phenomenon where models demonstrate mastery over the deterministic start and end regions but suffer a performance collapse in the complex intermediate sections. This degradation arises because standard training allows the massive volume of easy patterns to drown out the learning signals from difficult layouts. To address this, we propose \textbf{FocalOrder}, a framework driven by \textbf{Focal Preference Optimization (FPO)}. Specifically, FocalOrder employs adaptive difficulty discovery with exponential moving average mechanism to dynamically pinpoint hard-to-learn transitions, while introducing a difficulty-calibrated pairwise ranking objective to enforce global logical consistency. Extensive experiments demonstrate that FocalOrder establishes new state-of-the-art results on OmniDocBench v1.0 and Comp-HRDoc. Our compact model not only outperforms competitive specialized baselines but also significantly surpasses large-scale general VLMs. These results demonstrate that aligning the optimization with intrinsic structural ambiguity of documents is critical for mastering complex document structures. |
| title | FocalOrder: Focal Preference Optimization for Reading Order Detection |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.07483 |