Multimodal Planning with Complex Constraints (MPCC)
收藏资源简介:
MPCC 数据集是一个针对多模态大语言模型在复杂约束下的规划能力进行系统评估的基准数据集。数据集由三个现实世界任务组成:航班规划、日历规划和会议规划,每个任务都有逐渐增加的约束难度。数据集包含三种类型的复杂约束:预算约束、时间约束和空间约束。数据集由真实场景生成,并通过人工筛选确保约束的准确性和组合多样性。MPCC 的目标是促进多模态规划领域的研究,并为多模态大语言模型提供更具挑战性的评估框架。
The MPCC dataset is a benchmark dataset for systematically evaluating the planning capabilities of multimodal large language models under complex constraints. It consists of three real-world tasks: flight planning, calendar planning, and meeting planning, each with gradually increasing constraint difficulty. The dataset includes three types of complex constraints: budget constraints, temporal constraints, and spatial constraints. It is generated from real-world scenarios and manually screened to ensure the accuracy of constraints and the diversity of their combinations. The goal of MPCC is to promote research in the field of multimodal planning and provide a more challenging evaluation framework for multimodal large language models.
MPCC: 多模态复杂约束规划基准数据集
数据集概述
- 全称: Multimodal Planning with Complex Constraints (MPCC)
- 类型: 多模态规划基准测试集
- 应用领域: 多模态大语言模型(MLLMs)的复杂约束规划能力评估
- 收录会议: ACM MM 2025
核心特点
- 约束建模: 显式建模复杂约束条件
- 评估指标: 提出可行性(feasibility)和最优性(optimality)双指标评估体系
- 任务多样性: 覆盖航班/日历/会议等规划场景
- 约束类型: 包含空间/时间/预算等多类约束条件
数据集构成
- 领域划分:
- Flight Planning (航班规划)
- flight_plan_easy.parquet
- flight_plan_medium.parquet
- flight_plan_hard.parquet
- Calendar Planning (日历规划)
- Meeting Planning (会议规划)
- Flight Planning (航班规划)
评估方法
- 推荐工具: VLMEvalKit (https://github.com/open-compass/vlmevalkit)
- 评估对象: 支持13种开源/闭源MLLMs评估
获取方式
- HuggingFace地址: https://huggingface.co/datasets/jyyyyy67/MPCC
- 加载代码: python from datasets import load_dataset dataset = load_dataset("jyyyyy67/MPCC", data_files="[领域]/[文件名]")
联系方式
- 维护者: Yiyan Ji
- 邮箱: jiyiiiyyy@gmail.com
- 问题反馈: GitHub Issues




