Research map / Vision-language and multimodal / Multimodal LLMs
Beyond the Eye: Efficient Multimodal Reasoning Via Self-Regulated Implicit Visual Tools
Xiuwei Chen, Quanlin Chen, Wentao Hu, Zisheng Chen, Kun Xiang, Zehua Ma, Mingyang Zhang, Jianhua Han, Hanhui Li, Hang Xu, Xiaodan Liang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add implicit visual tools and BEE training to Qwen2.5-VL-7B | Qwen2.5-VL-7B · V* | accuracy: 76.4 -> 83.8 | Improved |
| add implicit visual tools to Qwen3-VL-4B | Qwen3-VL-4B · V* | accuracy: 80.6 -> 91.6 | Improved |
| add implicit visual tools to Qwen3-VL-8B | Qwen3-VL-8B · MME-Real-Lite | accuracy: 49.3 -> 59.4 | Improved |
| add implicit visual tools to Qwen3-VL-8B | Qwen3-VL-8B · POPE | accuracy: 87.6 -> 87.7 | No clear effect |
| add implicit visual tools to Qwen3-VL-4B | Qwen3-VL-4B · LogicVista | accuracy: 53.2 -> 53.4 | No clear effect |
| add MC-Reward (self-regulated tool use) instead of static reward (ablation) | BEE-7B · NTG | NTG: 6.2 -> 66.0 | Improved |
| add MC-Reward instead of static reward (ablation) | BEE-7B · system latency | avg. latency (s): 2.53 -> 1.37 | Improved |
| add MC-Reward instead of static reward (ablation) | BEE-7B · system throughput | throughput (sample/s): 0.88 -> 1.61 | Improved |
| use shared single token instead of generated implicit tokens (Replace) (ablation) | BEE-7B · V* overall | accuracy: 81.1 -> 83.8 | Improved |
| use target image instead of original image, vs unk tokens, vs text-only (ablation) | BEE-7B · HRBench | accuracy: largest drop on HRBench 8K when target image replaced by original | Improved |
| add adapter decoupling text and visual token optimization (ablation) | BEE-7B · V* overall | accuracy: 83.8 -> 84.2 | Improved |
| add adapter decoupling text and visual token optimization (ablation) | BEE-7B · HRBench-4K overall | accuracy: 75.4 -> 76.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DeepEyes — DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026) · better on 3 of 3 shared results
- DeepEyesV2 — DeepEyesV2: Toward Agentic Multimodal Model (ICLR 2026) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Don't Show Pixels, Show Cues: Unlocking Visual Tool Reasoning in Language Models via Perception Programs (CVPR 2026 · similarity 0.91)
- Latent Implicit Visual Reasoning (CVPR 2026 · similarity 0.90)
- Thinking with Programming Vision: Towards a Unified View for Thinking with Images (CVPR 2026 · similarity 0.90)
- Learn to Think: Improving Multimodal Reasoning through Vision-Aware Self-Improvement Training (ICML 2026 · similarity 0.90)
- MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods (NeurIPS 2026 · similarity 0.90)
- See It, Say It, Sorted: An Iterative Training-Free Framework for Visually-Grounded Multimodal Reasoning in LVLMs (CVPR 2026 · similarity 0.90)
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.
