EgoPlan-Bench2
收藏资源简介:
EgoPlan-Bench2是由香港大学和腾讯PCG ARC实验室创建的多模态大语言模型规划基准,旨在评估模型在多种真实世界场景中的规划能力。该数据集包含1,321个高质量的多选题问答对,覆盖了工作、日常生活、爱好和娱乐四大领域,共24个详细场景。数据集通过半自动化的过程构建,利用第一人称视角的视频,结合手动验证,确保数据的真实性和可靠性。EgoPlan-Bench2主要用于评估和提升多模态大语言模型在复杂环境中的任务规划能力,旨在解决现实世界中的多样化问题。
EgoPlan-Bench2 is a multimodal large language model planning benchmark created by the University of Hong Kong and Tencent PCG ARC Lab, which aims to evaluate the planning capabilities of models across various real-world scenarios. This dataset contains 1,321 high-quality multiple-choice question-answer pairs, covering four major domains: work, daily life, hobbies and entertainment, with a total of 24 detailed scenarios. The dataset is constructed through a semi-automated process, leveraging first-person perspective videos combined with manual verification to ensure the authenticity and reliability of the data. EgoPlan-Bench2 is primarily used to evaluate and enhance the task planning capabilities of multimodal large language models in complex environments, with the goal of solving diverse real-world problems.

- 1EgoPlan-Bench2: A Benchmark for Multimodal Large Language Model Planning in Real-World Scenarios香港大学, 腾讯PCG ARC实验室 · 2024年



