Research map / Vision-language and multimodal / Multimodal LLMs
Context-Aware Reasoner: Enhancing Contextual Reasoning in Multimodal Large Language Models
Zhe Zheng, Wenqi Zhang, Xiaohe Zhou, Guiyang Hou, Xinyu Mei, Tianyang Mao, Anni PIAO, Shengpei Jiang, Yongliang Shen, Yueting Zhuang, Weiming Lu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces ContextReasonV-Bench to evaluate contextual reasoning in multimodal LLMs, identifying two failure modes, and proposes Context-Aware Reasoner to address them (details beyond the abstract unavailable).
Problem
MLLMs are evaluated mainly on reasoning over internalized knowledge, while reasoning grounded in information present in the context (contextual reasoning) is overlooked.
Approach
Constructs a benchmark specifically for visual contextual reasoning and uses its failure analysis to motivate a context-aware reasoning approach; the specific method is not described in the available abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add context-aware reward to vanilla GRPO (ablation) | CAR RL stage · ContextReasonV-Bench Overall | accuracy (%): 33.20 -> 34.59 | Improved |
| add hierarchical advantage estimation with context augmentation (ablation) | CAR RL stage · ContextReasonV-Bench Overall | accuracy (%): 34.59 -> 38.14 | Improved |
| add context semantic augmentation (without HAE) (ablation) | CAR RL stage · ContextReasonV-Bench Overall | accuracy (%): 34.59 -> 36.74 | Improved |
| add HAE advantage estimation on top of augmented contexts (ablation) | CAR RL stage · ContextReasonV-Bench Overall | accuracy (%): 36.74 -> 38.14 | Improved |
| add SFT on 388 analyze-then-solve samples | Qwen2.5-VL-7B · ContextReasonV-Bench Overall | accuracy (%): 16.05 -> 27.68 | Improved |
| add RL stage after SFT | CAR-SFT · ContextReasonV-Bench Overall | accuracy (%): 27.68 -> 38.14 | Improved |
| SFT on out-of-distribution format task | CAR-SFT · Visual Logical Transformation | accuracy (%): 25.40 -> 23.01 | Hurt |
| SFT on out-of-distribution format task | CAR-SFT · Symbolic Transformation Learning | accuracy (%): 35.00 -> 15.00 | Hurt |
| rule-based math RL (Vision-R1-7B) on base model | Qwen2.5-VL-7B · ContextReasonV-Bench Overall | accuracy (%): 16.05 -> 14.85 | No clear effect |
| rule-based math RL (MM-Eureka-7B) on base model | Qwen2.5-VL-7B · ContextReasonV-Bench Overall | accuracy (%): 16.05 -> 16.38 | No clear effect |
| apply CAR training | Qwen2.5-VL-7B · General vision tasks average | accuracy (%): 45.58 -> 49.60 | Improved |
| apply CAR training | Qwen2.5-VL-7B · MMMU | accuracy (%): 53.78 -> 55.56 | Improved |
Limitations
- Full text unavailable; only the abstract was provided
Experimental setup
Benchmarks: ContextReasonV-Bench
Closest papers by meaning
- Open Vision Reasoner: Transferring Linguistic Cognitive Behavior for Visual Reasoning (NeurIPS 2025 · similarity 0.92)
- Activating Visual Context and Commonsense Reasoning Through Masked Prediction in VLMs (AAAI 2026 · similarity 0.91)
- GThinker: Towards General Multimodal Reasoning via Cue-Guided Rethinking (CVPR 2026 · similarity 0.90)
- VisuRiddles: Fine-grained Perception is a Primary Bottleneck for Multimodal Large Language Models in Abstract Visual Reasoning (ICLR 2026 · similarity 0.90)
- Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning of Vision Language Models (NeurIPS 2025 · similarity 0.89)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
