SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows

查看原文 / View Original 编辑 / Edit

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.10563 (cs) [Submitted on 1 Oct 2026] Title: SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows Authors: Albert Gao, Bing Xue, Andrea Zanette.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.10563 (cs) [提交于 2026 年 10 月 1 日] 标题：SLVR：通过类人推理流实现的结构化潜在视觉推理作者：Albert Gao, Bing Xue, Andrea Zanette。

Abstract: Multimodal large language models (MLLMs) often answer visual reasoning questions by relying on linguistic priors rather than task-relevant visual evidence. Textual chain-of-thought reasoning can partially mitigate this issue by encouraging models to decompose visual questions into intermediate evidence-seeking steps, but generating these steps autoregressively increases inference cost.

摘要：多模态大语言模型（MLLMs）在回答视觉推理问题时，往往依赖语言先验而非任务相关的视觉证据。文本思维链推理可以通过鼓励模型将视觉问题分解为中间的证据搜寻步骤来部分缓解这一问题，但自回归生成这些步骤会增加推理成本。

Latent reasoning avoids explicit rationale generation, but existing approaches provide limited control over what intermediate states encode, making it difficult to impose separate supervision for planning, grounding, and evidence selection. We propose Structured Latent Visual Reasoning (SLVR), a training framework that bridges explicit chain-of-thought and latent reasoning by organizing multimodal reasoning into typed latent stages for planning, grounding, evidence selection, and reasoning integration.

潜在推理避免了显式的逻辑生成，但现有方法对中间状态编码内容的控制有限，难以对规划、定位和证据选择实施单独的监督。我们提出了结构化潜在视觉推理（SLVR），这是一个通过将多模态推理组织为规划、定位、证据选择和推理整合等类型化潜在阶段，从而连接显式思维链与潜在推理的训练框架。

SLVR first trains the model to rely on the image by masking answer-revealing text and contrasting the correct answer with visually plausible distractors. It then organizes reasoning into latent stages for planning, grounding, evidence selection, and integration, supervising each stage with the corresponding signal: plans, boxes, visual evidence, and final rationales.

SLVR 首先通过掩盖揭示答案的文本并将正确答案与视觉上合理的干扰项进行对比，训练模型依赖图像。随后，它将推理组织为规划、定位、证据选择和整合的潜在阶段，并使用相应的信号（计划、边界框、视觉证据和最终逻辑）对每个阶段进行监督。

This gives latent reasoning an explicit functional structure while avoiding generated textual chains at inference time. Built on Qwen2.5-VL-7B, SLVR improves consistently across multimodal reasoning benchmarks, with absolute gains of +9.4 on MMVP and +14.2 on BLINK Relation, as well as improvements on V*, MathVista, and ChartQA.

这赋予了潜在推理明确的功能结构，同时避免了在推理时生成文本链。基于 Qwen2.5-VL-7B 构建的 SLVR 在多模态推理基准测试中表现出持续提升，在 MMVP 上绝对提升了 +9.4，在 BLINK Relation 上提升了 +14.2，并在 V*、MathVista 和 ChartQA 上也取得了进步。

These results suggest that structured latent supervision can improve fine-grained visual reasoning without the decoding overhead of textual CoT. Project page is available here.

这些结果表明，结构化潜在监督可以在没有文本思维链解码开销的情况下，改善细粒度的视觉推理。项目页面可在此处访问。