洞穴的寓言：基于测量的视觉-语言学习
Allegory of the Cave: Measurement-Grounded Vision-Language Learning
AI总结：
本文研究了视觉-语言模型在更接近底层相机测量的视觉接口下，基于测量的视觉-语言学习是否能提升性能。PRISM-VL模型结合RAW数据、相机条件接地和曝光括号监督聚合，通过高质量数据集和基准测试，提升了BLEU、ROUGE-L和LLM-Judge的准确率。
摘要：
视觉-语言模型通常基于经过ISP处理的RGB图像进行推理，尽管RGB渲染可能在推理前裁剪、抑制或量化传感器证据。我们研究当视觉接口更接近底层相机测量时， grounding 是否能提升性能。我们提出了测量 grounded 的视觉-语言学习，并将其实例化为PRISM-VL，该模型结合了RAW衍生的Meas.-XYZ输入、相机条件接地和曝光括号监督聚合，以将监督从RGB代理转移到测量域观察。使用一个质量控制的15万条指令微调数据集和一个针对低光、HDR、可见性敏感和幻觉敏感情况的验证基准，PRISM-VL-8B达到了0.6120 BLEU、0.4571 ROUGE-L和82.66%的LLM-Judge准确率，比RGB Qwen3-VL-8B基线提高了+0.1074 BLEU、+0.1071 ROUGE-L和+4.46个百分点。这些结果表明，部分VLM grounding误差源于RGB渲染过程中信息的丢失，而保留测量域证据可以提高多模态推理。
英文摘要：
Vision-language models typically reason over post-ISP RGB images, although RGB rendering can clip, suppress, or quantize sensor evidence before inference. We study whether grounding improves when the visual interface is moved closer to the underlying camera measurement. We formulate measurement-grounded vision-language learning and instantiate it as PRISM-VL, which combines RAW-derived Meas.-XYZ inputs, camera-conditioned grounding, and Exposure-Bracketed Supervision Aggregation for transferring supervision from RGB proxies to measurement-domain observations. Using a quality-controlled 150K instruction-tuning set and a held-out benchmark targeting low-light, HDR, visibility-sensitive, and hallucination-sensitive cases, PRISM-VL-8B reaches 0.6120 BLEU, 0.4571 ROUGE-L, and 82.66\% LLM-Judge accuracy, improving over the RGB Qwen3-VL-8B baseline by +0.1074 BLEU, +0.1071 ROUGE-L, and +4.46 percentage points. These results suggest that part of VLM grounding error arises from information lost during RGB rendering, and that preserving measurement-domain evidence can improve multimodal reasoning.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
