不重采图也不多轮搜索，用中间层路由已有视觉证据提升高分辨率图文问答性能。
Thinking Once Is Enough: Intermediate-Layer Evidence Routing for High-Resolution VQA
- 在视觉编码中间层重建问题相关的注意力，直接路由关键信息到后续层。
- 在多个基准上平均提分3.1~9.9点，峰值内存降低约4GB，推理速度提升97.2%。
- 无需训练、单次图像处理，适合部署于资源受限的高分辨率图文问答场景。
高分辨率视觉问答（HR-VQA）常被视为证据获取不足的问题，即多模态大模型需通过裁剪、重编码或多轮搜索重复检查图像。我们指出这一观点不完整：许多情况下，细粒度证据已在视觉编码中保留，并在中间层路由窗口内可识别且具影响力，但随后被稀释。为此，我们提出Thinking-Once，一种无需训练、仅一次视觉输入的证据路由方法：在中间层重建问题相关的注意力，保留核心实体与紧凑背景上下文，并将其路由至后层，无需额外视觉编码。在五个基线模型上，Thinking-Once始终优于或匹配原基线，在V$^*$Bench、HRBench-4K和HRBench-8K上分别提升+3.1、+3.0、+2.7分，平均峰值内存减少约4GB。在Qwen2.5-VL-7B上，三基准提升+9.9、+4.6、+5.5分，跨基准均值从72.5升至79.1。使用ZwZ-8B时，跨基准均值达82.7。相比11个开源基线，其在三个基准平均分及整体均值上均最优或并列最优；例如对比DeepScan，V$^*$Bench推理时间减少97.2%，跨基准均值从77.8提升至79.1。结果表明，通过路由已编码证据即可改进HR-VQA，无需反复获取新视觉输入。代码见附录。
原文摘要 · Abstract (English)
High-resolution visual question answering (HR-VQA) is often treated as a problem of insufficient evidence acquisition, where failing multimodal large language models must inspect images again through cropping, re-encoding, or multi-round search. We show that this view is incomplete: in many cases, fine-grained evidence has already survived visual encoding and become identifiable and influential within an intermediate-layer routing window, but is later diluted before answer generation. We propose Thinking-Once, a \textbf{training-free, single-visual-pass} evidence-routing method that reconstructs question-conditioned attention at this window, preserves core entity tokens and compact background context, and routes this evidence to later layers without extra visual encoding. Across five base models, Thinking-Once consistently improves or matches the corresponding base setting, increasing the average scores on V$^*$Bench, HRBench-4K, and HRBench-8K by \textit{+3.1}, \textit{+3.0}, and \textit{+2.7} points while reducing the average peak memory by about 4,GB. On Qwen2.5-VL-7B, it improves the three benchmarks by \textit{+9.9}, \textit{+4.6}, and \textit{+5.5} points, raising the cross-benchmark mean from 72.5 to 79.1. With the ZwZ-8B base model, Thinking-Once reaches a mean score of 82.7. Against 11 open-source HR-VQA baselines, it obtains the best or tied-best score on all three benchmark averages and the best overall mean; for example, compared with DeepScan, it reduces V$^*$Bench inference time by \textbf{97.2\%} while improving the cross-benchmark mean from 77.8 to 79.1. These results show that HR-VQA can be improved by routing already encoded evidence rather than repeatedly acquiring new visual inputs. Code is available in the appendix.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。