外观
论文HF Daily Papers · arXiv 2610.37037昨天
ZwZ2框架通过证据连接数据合成提升具身空间感知与长上下文理解
本文由 AI 根据公开内容整理，可能有误，请以原文为准：Zooming without Zooming v2: From Seeing Details to Connecting Evidence for Generalizable Multimodal Perception（HF Daily Papers · arXiv 2610.37037）
核心内容
针对具身智能中空间感知与多模态长上下文理解训练数据合成难的问题，研究团队提出EvidenceConnect框架及ZwZ2模型。该工作洞察到两类任务共享“连接证据”底层能力，即识别视觉事实、绑定实体并恢复关系。通过构建非特定下游基准的合成数据并采用RLVR训练，ZwZ2在未针对特定测试集优化的情况下，显著提升了细粒度感知、具身空间推理及通用视觉推理能力，验证了强化共享基础能力对多模态感知泛化的有效性。
技术实现上，EvidenceConnect利用单张图像相关区域的缩放视图，自动生成需比较、聚合或链式推理的视觉证据问题。为确保数据质量，仅保留独立VLM达成一致的答案，并通过区域消融测试剔除答案不依赖选定区域的无效样本。基座VLM基于此高质量合成数据进行RLVR训练获得ZwZ2。实验表明，这种针对“证据连接”能力的监督信号能有效迁移至具身Agent任务与长上下文理解场景，解决了 targeted 数据获取成本高的问题。
关键看点
- 证据连接机制：将空间感知与长上下文理解解耦为识别视觉事实、绑定实体及恢复关系的共享底层能力，降低定向数据合成门槛
- 数据合成策略：基于单图区域缩放视图生成比对/聚合/链式推理问答，利用独立VLM共识与区域消融测试双重过滤低质样本
- RLVR训练范式：基座VLM通过强化学习验证推理（RLVR）在合成数据上微调，无需下游Benchmark定制即可实现能力迁移
- 具身任务泛化：ZwZ2在非针对性训练下提升具身空间推理与Agent任务表现，证实基础感知能力增强对复杂任务的支撑作用
工程与产业观察
该研究为具身智能数据工程提供了新范式：与其昂贵地采集特定场景数据，不如通过结构化合成强化“证据连接”等原子能力。结合RLVR与严格的质量过滤机制，低成本合成数据即可驱动VLM在空间推理与长上下文理解上的泛化，对解决具身Agent训练数据瓶颈具有直接参考价值。
英文摘要 (Abstract)
Multimodal perception forms the foundation of multimodal intelligence. In particular, spatial perception and multimodal long-context understanding are two pivotal tasks that underpin embodied AI and multimodal agents. However, synthesizing targeted, high-quality training data for these tasks remains difficult and costly. Our key insight is that these tasks share an underlying capability, which we call connecting evidence: identifying relevant visual facts, binding them to the correct entities, recovering their relations, and composing them into an answer. Crucially, targeted training data are far easier to synthesize for this capability than for these tasks themselves. To strengthen this capability, we propose EvidenceConnect, an easy-to-implement data synthesis framework that uses zoomed views of related regions within a single image to generate questions requiring comparison, aggregation, or chaining of visual evidence. To ensure quality, a reference answer is retained only when independent VLMs agree on it, and region-ablation tests discard questions whose answer survives the removal of selected regions. We then train base VLMs on the resulting data with RLVR, obtaining Zooming without Zooming v2 (ZwZ2). Although the synthesized data are not tailored to any downstream benchmark, ZwZ2 improves fine-grained and spatial perception, with further gains on embodied spatial reasoning, multimodal long-context understanding, multimodal agent tasks, and general visual reasoning. These results establish EvidenceConnect as an effective framework for supervising evidence connection and suggest that strengthening this shared capability leads to more generalizable multimodal perception.
Hugging Face Daily Papers 热度：18 upvotes 作者：Lai Wei, Yatai Li, Liangbo He, Jun Lan, Renjie Gu, Linhan Cao, Junkai Lin, Huijia Zhu, Weiqiang Wang, Linghe Kong, Yue Wang, Zhuosheng Zhang, Weiran Huang
来源
本页事实对应的公开页面，以原文为准。