arXiv cs.AI· Gregory Kang Ruey Lau, Minh Huynh Nguyen, Bryan Kian Hsiang Low·· 10 小时前AI 评分42
TIGER：弥合多模态大模型的视觉推理接入差距
Seeing Isn't Enough: Unlocking Visual Reasoning by Bridging the Reasoning-Access Gap with TIGER
AI 导读
研究提出 TIGER（Text-to-Image Gap-targeted training for Enhanced Reasoning），通过挖掘模型在文本题上成功、在对应渲染图像上失败的合成模态反事实样本，并用 GRPO 结合 SFT、DPO 变体对图像输入进行后训练。
正文 · 原文
Abstract:Multimodal Large Language Models (MLLMs) inherit strong textual reasoning capabilities from their language backbones and text pretraining, but often fail to apply the same level of competency to visual inputs. We show that this degradation cannot be attributed to only perception failures, by evaluating models on semantically-equivalent text and rendered-text image inputs: models can accurately read text in the images and recover near-text performance when provided their own transcriptions, but just cannot reason well on the images directly. We characterize this discrepancy as a modality reasoning-access gap, where models fail to route perceived visual content into existing reasoning machinery accessible with text. To bridge this, we propose TIGER (Text-to-Image Gap-targeted training for Enhanced Reasoning), which first mines synthetic modality counterfactuals where a model succeeds on a text problem but fails when provided the corresponding rendered image, and post-trains the model on the image input via Group Relative Policy Optimization (GRPO), along with SFT and DPO variants. Importantly, although trained exclusively on synthetic rendered-text images with no multimodal annotations, TIGER results in model performance improvements across diverse real-world visual benchmarks including geometric diagrams (MathVerse), multi-disciplinary sciences (EMMA, MMMU), complex charts (ChartQAPro), and embodied spatial reasoning (RoboSpatial-Home). TIGER also complements existing state-of-the-art RLVR visual reasoning methods, yielding further performance gains. Analysis involving reasoning-subspace activation and activation patching provide support indicating that TIGER helps to align visual-to-reasoning pathways toward text reasoning states. Our implementation is available at this https URL.
| Comments: | Accepted at NeurIPS 2026. Gregory and Minh contributed equally |
| Subjects: | Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) |
| Cite as: | arXiv:2611.10552 [cs.AI] |
| (or arXiv:2611.10552v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.10552 arXiv-issued DOI via DataCite |
Submission history
From: Huynh Minh Nguyen [view email]
[v1]
Sat, 7 Nov 2026 07:41:50 UTC (6,021 KB)
[v2]
Tue, 10 Nov 2026 03:36:09 UTC (6,021 KB)
来源：arXiv cs.AI · arxiv.org