MMR
收藏资源简介:
MMR数据集是由韩国科学技术院(KAIST)的研究团队构建的一个大规模推理分割数据集。该数据集包含194,398个复杂且隐晦的指令-问题对,涵盖了多目标、对象级和部分级推理。数据集基于现有的PACO-LVIS图像-掩膜集,通过GPT-4V API生成复杂的指令。MMR数据集的特点是能够处理多个对象和多样的部分,在单个查询中提供对象和部分的信息,支持多样化和上下文感知的交互。
The MMR dataset is a large-scale reasoning segmentation dataset constructed by a research team from the Korea Advanced Institute of Science and Technology (KAIST). It contains 194,398 complex and ambiguous instruction-question pairs, covering multi-object, object-level, and part-level reasoning. Built upon the existing PACO-LVIS image-mask set, complex instructions for the dataset are generated via the GPT-4V API. The MMR dataset is characterized by its ability to handle multiple objects and diverse parts, provide information about both objects and parts in a single query, and support diverse and context-aware interactions.

- 1MMR: A Large-scale Benchmark Dataset for Multi-target and Multi-granularity Reasoning Segmentation韩国科学技术院(KAIST) · 2025年



