以编程视觉思考：迈向用图像思考的统一视角
Thinking with Programming Vision: Towards a Unified View for Thinking with Images
把「生成代码调用任意图像操作」作为统一工具接口，替代固定工具注册表，并用稠密过程奖励的强化学习训练多轮工具组合与运行时错误恢复；做视觉推理、工具增强 VLM 或多模态 agent 的工作可直接复用该框架与配套的朝向变化鲁棒性基准。
原文依据 we propose CodeVision, a flexible and scalable "code-as-tool" framework where the model generates code as a universal interface to invoke any image operation
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论能够「用图像思考」的多模态大语言模型（MLLM）可以交互式地使用工具对视觉输入进行推理，但现有方法往往依赖一组范围狭窄的工具，其现实必要性与可扩展性都较为有限。在本工作中，我们首先揭示了一个关键且此前被忽视的弱点：即使是最优的 MLLM 也出人意料地脆弱，在仅发生简单朝向变化的图像上就表现出显著的性能下降，这凸显了对更鲁棒的基于工具的推理的需求。为解决这一问题，我们提出 CodeVision，一个灵活且可扩展的「代码即工具」框架，模型生成代码作为通用接口来调用任意图像操作，从而超越固定的工具注册表。我们采用两阶段方法训练模型：首先在一个为复杂多轮工具组合与错误恢复而精心构建的高质量数据集上进行监督微调（SFT），随后使用稠密过程奖励函数进行强化学习（RL），以鼓励策略性且高效的工具使用。为促进这一研究，我们构建了新的 SFT 与 RL 数据集，并引入一套具有挑战性的新基准套件，旨在严格评估对朝向变化的鲁棒性以及多工具推理能力。在 Qwen2.5-VL 与 Qwen3-VL 系列上的实验表明，我们的方法显著提升了模型性能，并催生出灵活的工具组合、高效的链式执行以及基于运行时反馈的鲁棒错误恢复等涌现能力。代码已开源：https://github.com/ByteDance-BandAI/CodeVision。
英文原文摘要 1570 字符
Multimodal large language models (MLLMs) that "think with images" can interactively use tools to reason about visual inputs, but current approaches often rely on a narrow set of tools with limited real-world necessity and scalability. In this work, we first reveal a critical and previously overlooked weakness: even state-of-the-art MLLMs are surprisingly brittle, showing significant performance degradation on images with simple orientation changes, underscoring the need for more robust tool-based reasoning. To address this, we propose CodeVision, a flexible and scalable "code-as-tool" framework where the model generates code as a universal interface to invoke any image operation, moving beyond fixed tool registries. We train our model using a two-stage methodology, beginning with Supervised Fine-Tuning (SFT) on a high-quality dataset curated for complex, multi-turn tool composition and error recovery, followed by Reinforcement Learning (RL) with a dense process reward function to encourage strategic and efficient tool use. To facilitate this research, we construct new SFT and RL datasets and introduce a challenging new benchmark suite designed to rigorously evaluate robustness to orientation changes and multi-tool reasoning. Experiments on Qwen2.5-VL and Qwen3-VL series show that our approach significantly improves model performance and fosters emergent capabilities such as flexible tool composition, efficient chained execution, and robust error recovery from runtime feedback. Code is available at https://github.com/ByteDance-BandAI/CodeVision.