论文

研究提出GIB-200K基准:VLM面对检索噪声时并不像理想观察者那样权衡文本与图像

The Illusion of Grounding: Do Vision-Language Models Weigh Retrieved Text and Images Like an Ideal Observer?

精选理由

做检索增强或多模态应用的人该看看:论文实测 LLaVA 偏信检索文本、InternVL3 偏信图像,文档说谎时模型会放弃高达 71% 的正确答案。200K 条基准和代码全开源,可以直接拿来测自己的系统。

arXiv 论文(编号 2611.23127)提出 GIB-200K 基准,包含 200,000 个基于 COCO 和 Visual Genome 构建的图像-问题条目,覆盖 1-20 篇难负检索噪声文档、高斯与运动模糊、反事实翻转对等受控压力条件。论文从贝叶斯线索组合推导出可检验预测,形式化定义了 Knowledge Inertia(文本被过度加权)与 Sensory Capture(相反情形),并用测量误差校正回归估计权重,另以 Grounding Illusion Score 衡量干净准确率中有多少并未真正基于图像。

正文

在 2,000 条冻结评估集上,4 个开源模型在干净输入下准确率为 92.5%-94.2%,但在不利条件下跌至 41.3%-59.9%。两个 LLaVA 模型的文本/图像权重比 w_T/w_I 分别为 1.229 和 1.322(偏向文本),InternVL3 则为 0.680(偏向图像);当检索文档说谎时,34.8%-71.4% 原本正确的答案被放弃。

论文还比较了两种缓解方案:免训练的一致性检查(对比有无检索的回答)在误导文档下用覆盖率换准确率,而图像严重退化时回退到图像则失效。所有逐条输出、代码和复算审计已开源,RAG 系统开发者可直接据此测试自己的模型组合策略。

原文 · arXiv cs.LG

The Illusion of Grounding: Do Vision-Language Models Weigh Retrieved Text and Images Like an Ideal Observer?

Retrieval-augmented vision-language models (VLMs) answer from two sources of evidence that can each fail: the retrieved text can be wrong or noisy, and the image can be degraded or contradicted. We introduce GIB-200K, 200,000 image-question items built from COCO and Visual Genome, each expanded into controlled stress conditions: hard-negative retrieval noise (1-20 documents), visual ambiguity (Gaussian and motion blur, low-resource crops) and cross-modal conflict with counterfactual flip pairs. Instead of only measuring degradation, we ask how models combine the two modalities. From ideal Bayesian cue combination we derive a testable prediction that yields formal definitions of Knowledge Inertia (text over-weighted relative to the model's own image evidence) and Sensory Capture (the reverse), and we estimate the weights with a measurement-error-corrected regression whose recovery of known weights we verify. A counterfactual flip decomposition separates modality bias from concept priors, and a Grounding Illusion Score measures how much clean accuracy is not grounded in the image. We evaluate 4 open models on one frozen, seeded evaluation split of 2,000 GIB-200K items. All models are accurate on clean inputs (92.5%-94.2%) but adverse accuracy falls to 41.3%-59.9%. The two LLaVA models over-weight the text (Knowledge Inertia, w_T/w_I = 1.229 and 1.322), whereas InternVL3 favours the image (w_T/w_I = 0.680); between 34.8% and 71.4% of the clean answers that are correct are abandoned when the document lies. A training-free consistency check that compares the answer with and without retrieval trades coverage for accuracy under misleading documents, while falling back to the image fails when the image is heavily degraded. All per-instance outputs, code, and an audit that recomputes every reported open-model number are released.

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:VLM、RAG、GIB-200K、LLaVA,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.LG」(@Basavaraj R M, Harshal Chandel),经 AITOP 多维评分管线结构化提炼。