CogDoc: Towards Unified thinking in Documents
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914201553862656 |
|---|---|
| author | Xu, Qixin Wang, Haozhe Liu, Che Lin, Fangzhen Chen, Wenhu |
| author_facet | Xu, Qixin Wang, Haozhe Liu, Che Lin, Fangzhen Chen, Wenhu |
| contents | Current document reasoning paradigms are constrained by a fundamental trade-off between scalability (processing long-context documents) and fidelity (capturing fine-grained, multimodal details). To bridge this gap, we propose CogDoc, a unified coarse-to-fine thinking framework that mimics human cognitive processes: a low-resolution "Fast Reading" phase for scalable information localization,followed by a high-resolution "Focused Thinking" phase for deep reasoning. We conduct a rigorous investigation into post-training strategies for the unified thinking framework, demonstrating that a Direct Reinforcement Learning (RL) approach outperforms RL with Supervised Fine-Tuning (SFT) initialization. Specifically, we find that direct RL avoids the "policy conflict" observed in SFT. Empirically, our 7B model achieves state-of-the-art performance within its parameter class, notably surpassing significantly larger proprietary models (e.g., GPT-4o) on challenging, visually rich document benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_12658 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | CogDoc: Towards Unified thinking in Documents Xu, Qixin Wang, Haozhe Liu, Che Lin, Fangzhen Chen, Wenhu Computer Vision and Pattern Recognition Current document reasoning paradigms are constrained by a fundamental trade-off between scalability (processing long-context documents) and fidelity (capturing fine-grained, multimodal details). To bridge this gap, we propose CogDoc, a unified coarse-to-fine thinking framework that mimics human cognitive processes: a low-resolution "Fast Reading" phase for scalable information localization,followed by a high-resolution "Focused Thinking" phase for deep reasoning. We conduct a rigorous investigation into post-training strategies for the unified thinking framework, demonstrating that a Direct Reinforcement Learning (RL) approach outperforms RL with Supervised Fine-Tuning (SFT) initialization. Specifically, we find that direct RL avoids the "policy conflict" observed in SFT. Empirically, our 7B model achieves state-of-the-art performance within its parameter class, notably surpassing significantly larger proprietary models (e.g., GPT-4o) on challenging, visually rich document benchmarks. |
| title | CogDoc: Towards Unified thinking in Documents |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2512.12658 |