EgoPlan-Bench
收藏资源简介:
EgoPlan-Bench是一个专为评估自我中心具身规划中的多模态大语言模型(MLLMs)而设计的基准数据集。该数据集由腾讯人工智能实验室创建,包含从真实世界视频中提取的现实任务,涉及与数百种不同对象的交互和来自不同场景的复杂视觉观察。数据集的创建过程涉及自动提取任务目标,并构造基于这些目标的多选题。EgoPlan-Bench旨在解决MLLMs在实际环境中作为具身任务规划者的能力评估问题,特别是在需要复杂视觉理解和任务规划的场景中。
EgoPlan-Bench is a benchmark dataset specifically designed for evaluating multimodal large language models (MLLMs) in egocentric embodied planning. Created by Tencent AI Lab, this dataset contains realistic tasks extracted from real-world videos, involving interactions with hundreds of distinct objects and complex visual observations from diverse scenarios. The dataset construction process involves automatically extracting task objectives and constructing multiple-choice questions based on these objectives. EgoPlan-Bench aims to address the problem of evaluating the capabilities of MLLMs as embodied task planners in real-world environments, especially in scenarios requiring complex visual understanding and task planning.

- 1EgoPlan-Bench: Benchmarking Egocentric Embodied Planning with Multimodal Large Language Models腾讯人工智能实验室 · 2024年



