论文

ViRA 用 CrossRPO 训练视觉智能体,读图而非靠常识猜

ViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy Optimization

精选理由

这篇论文的思路很聪明:只改图里一个小物件让答案变化,模型要解对整组题就必须真的去读图。只用 3500 条数据,27B 版本就在多数基准上超过了 Claude Haiku 5.5 和 GPT-6-Luna。做视觉理解或文档解析的工程师可以直接借鉴对比集的构造方法。

论文提出 Cross-Input Relative Policy Optimization(CrossRPO),训练视觉智能体在 21 到 120 兆像素的超高分辨率图像上通过缩放局部区域来读证据作答,而不是依赖常识猜测。其机制是构建对比集:同一张图只在小区域内改动一个物体并使答案不同,只有真正读取证据才能解出整组问题,CrossRPO 以整组解答为奖励,作者并证明该方式不会强化基于常识的作答。团队据此构建了训练集 ViRA-Train 和评测集 ViRA-Bench,另用真实拍摄变化的 ViRA-Bench-Real 排除编辑痕迹。仅用约 3,500 条数据训练的 ViRA-9B 与 ViRA-27B 在 ViRA-Bench 最高分辨率下取得 63.3% 的整组准确率,高于 GRPO 的 59.3%;ViRA-9B 是同尺寸开源模型中最准确的,ViRA-27B 在多数基准上超过以最高思考档运行的 Claude Haiku 5.5 和 GPT-6-Luna。

正文

对开发者而言,这项工作表明 RL 奖励设计而非数据规模是关键:3,500 条对比集数据即可超过闭源模型的表现。对做文档理解、卫星图像或工业质检的团队,ViRA-Bench 提供了一个区分'读证据'与'猜答案'的评测工具,CrossRPO 的对比集构造方法可以迁移到其他需要精读视觉细节的场景。

原文 · arXiv cs.AI

ViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy Optimization

Visual agents answer questions about ultra-high-resolution photos by zooming into the relevant regions. However, existing agents often answer from common sense instead of reading the evidence. Reinforcement learning (RL) on the final answer cannot correct this, because a guess earns the same reward as reading. We propose Cross-Input Relative Policy Optimization (CrossRPO), which trains agents on contrast sets. A contrast set asks one question about several versions of an image that differ only in a small region with the evidence and have different answers, so only reading the evidence solves the whole set. CrossRPO rewards solving the whole set, and we prove that it does not reinforce a set answered from common sense. We build contrast sets for training (ViRA-Train) and evaluation (ViRA-Bench) by editing one small object in images of 21 to 120 megapixels, and we photograph further sets with real changes (ViRA-Bench-Real) to exclude editing traces. We train two Visual Reasoning Agents, ViRA-9B and ViRA-27B, with CrossRPO on only about 3,500 items of ViRA-Train each. CrossRPO achieves 63.3% set accuracy on ViRA-Bench at the maximum resolution, against 59.3% for GRPO. ViRA-9B is the most accurate open model of its size on all benchmarks, and ViRA-27B outperforms Claude Haiku 5.5 and GPT-6-Luna, both at their highest thinking level, on most benchmarks.

产业坐标与信源核验

本条情报归属 AITOP「论文」核心追踪栏目。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:ViRA、视觉智能体、强化学习、多模态,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明:本文原始线索采编自「arXiv cs.AI」(@Huawei Lin, Jinting Liu, Tongyan Wang, Xuyang Chen, Ruihua Shi, Zhichao Xu, Jiaxin Qin, Guoshenghui Zhao, Baijian Yang, Yingjie Victor Chen, Tan Yu, Weijie Zhao),经 AITOP 多维评分管线结构化提炼。