kepeng/MeasL-150K-V1
收藏资源简介:
MeasL-150K-V1是官方PRSIMVL指令调优语料库,用于测量基础视觉推理。它与MeasL-Bench-V1评估协议对齐,并发布用于可重复的非商业研究。数据集包含152,517个指令调优示例和48,000个发布图像文件,主要用于多模态监督微调(SFT),以支持测量XYZ和元数据基础推理。该语料库已去除了与MeasL-Bench-V1的严格基准重叠,确保在指令调优中的可重复性,而非用于基准评估。数据集基于约700K自动注释候选构建,经过质量评分后保留518,433条记录,最终平衡了来源和问题结构,并打包为发布本地图像。
MeasL-150K-V1 is the official PRSIMVL instruction-tuning corpus for measurement-grounded visual reasoning. It is aligned with the MeasL-Bench-V1 evaluation protocol and released for reproducible non-commercial research. The dataset contains 152,517 instruction-tuning examples and 48,000 release image files, primarily used for multimodal SFT for Meas.-XYZ and metadata-grounded reasoning. It is decontaminated against MeasL-Bench-V1, intended for reproducible instruction tuning, not benchmark evaluation. The corpus was built from approximately 700K auto-annotated candidates, with 518,433 records after quality scoring, and the final examples were balanced by source and question structure, packaged with release-local images.




