kepeng/MeasL-Bench-V1

MeasL-Bench（全称：基于测量的视觉语言基准测试）是PRSIMVL项目发布的官方基准测试集，用于评估基于测量域输入的视觉语言模型。该数据集旨在测试一个核心问题：当RGB渲染丢失传感器证据时，基于测量域输入的模型是否能更可靠地恢复和推理。数据集强调低光照证据恢复、HDR和曝光敏感的基础、可见性和幻觉敏感查询，并包含标准RGB基础切片以确保覆盖范围。它包含2,183个示例，分为RAW测量路径和RGB对应版本，适用于评估RGB原生视觉语言模型和基于测量的方法（如Meas.-XYZ管道）。数据集按能力维度组织，包括颜色属性基础、数量基础、描述性场景基础等14个类别，并提供了评估协议和使用指南。MeasL-Bench (short for Measurement-grounded Language-Vision Benchmark) is the official held-out benchmark released with PRSIMVL for measurement-grounded vision-language evaluation. It is designed to test a core question: when RGB rendering loses sensor evidence, can a model grounded on measurement-domain input recover and reason more reliably? The benchmark emphasizes low-illumination evidence recovery, HDR and exposure-sensitive grounding, visibility-sensitive and hallucination-sensitive queries, plus standard RGB-sufficient grounding slices for coverage. It contains 2,183 examples, with matched RAW and RGB versions, and is suitable for evaluating both RGB-native VLMs and measurement-grounded methods (e.g., Meas.-XYZ pipelines). It is organized by capability dimensions such as Chromatic Attribute Grounding, Numerosity Grounding, and others, and includes evaluation protocols and usage instructions.