Bench4Merge
收藏资源简介:
Bench4Merge是由清华大学人工智能产业研究院创建的一个综合基准数据集,旨在评估自动驾驶车辆在密集交通中的合并能力。该数据集包含50,000个初始场景,来源于DJI密集交通数据集,并经过分类以增强多样性。数据集内容包括车辆状态、道路结构和交互行为,通过深度神经网络捕捉微观交互特征。创建过程中,数据集利用了大规模的密集交通数据进行训练,并结合了大型语言模型进行评估。Bench4Merge主要应用于自动驾驶领域的合并场景评估,旨在解决现有方法在复杂交互环境中的评估不足问题。
Bench4Merge is a comprehensive benchmark dataset created by the Institute for Artificial Intelligence Industry Research of Tsinghua University, aiming to evaluate the merging capability of autonomous vehicles in dense traffic scenarios. It contains 50,000 initial scenarios sourced from the DJI Dense Traffic Dataset, which are classified to enhance diversity. The dataset covers vehicle states, road structures and interactive behaviors, with microscopic interactive features captured via deep neural networks. During its development, the dataset leverages large-scale dense traffic data for training and incorporates large language models (LLMs) for evaluation. Bench4Merge is primarily applied to the evaluation of merging scenarios in the autonomous driving field, aiming to address the insufficient evaluation of existing methods in complex interactive environments.
数据集概述
基本信息
- 标题: Anonymized Repository - Anonymous GitHub
- 描述: 该数据集详情页面未提供具体描述信息。
其他信息
- 图标:
- Apple Touch Icon:
/favicon/apple-touch-icon.png - Favicon (32x32):
/favicon/favicon-32x32.png - Favicon (16x16):
/favicon/favicon-16x16.png
- Apple Touch Icon:
- 网站配置:
- Manifest:
/favicon/site.webmanifest - Shortcut Icon:
/favicon.ico - MS Application Config:
/favicon/browserconfig.xml - Theme Color:
#ffffff
- Manifest:
- 样式表:
- 主要样式表:
/css/all.min.css - 代码高亮样式表:
/css/prism.css
- 主要样式表:
- 字体:
- 使用字体:
Nunito(400, 700, 800)
- 使用字体:




