cocreator-driving-scene
收藏资源简介:
CoCreator Driving Scene 是一个驾驶场景因果推理数据集,由 CoCreator 工具自动构建而成。该数据集旨在支持自动驾驶和场景理解相关任务,特别是因果推理、视觉问答以及图像到文本生成。数据集包含 1227 个样本,每个样本由两部分组成:一是按时间顺序排列的帧序列(视频帧),每帧图像的分辨率为 1600×900,以 PIL.Image 列表形式存储;二是对应的因果描述文本(causal_text),以字符串形式提供,用于解释帧序列中的因果事件。每个样本还具有唯一的字符串 ID(从 0001 到 1227)。数据集为单语英语,采用 MIT 许可证,注释由机器生成,数据规模在 1K 到 10K 之间。它适用于视觉问答、图像字幕生成、因果推理和自动驾驶场景分析等任务。用户可以通过 HuggingFace datasets 库加载数据集,并浏览 HTML 预览文件查看数据示例。
CoCreator Driving Scene is a causal reasoning dataset for driving scenarios, automatically constructed using the CoCreator tool. This dataset is designed to support tasks related to autonomous driving and scene understanding, particularly causal reasoning, visual question answering (VQA), and image-to-text generation. The dataset comprises 1227 samples, each with two components: 1) a temporally ordered frame sequence (video frames), where each frame has a resolution of 1600×900 and is stored as a list of PIL.Image objects; 2) the corresponding causal description text (causal_text), a string that explicates the causal events within the frame sequence. Each sample also features a unique string ID ranging from 0001 to 1227. This dataset is monolingual in English, licensed under the MIT License, with machine-generated annotations, and its total data size ranges from 1K to 10K. It is suitable for a range of tasks including visual question answering (VQA), image captioning, causal reasoning, and autonomous driving scenario analysis. Users can load this dataset via the Hugging Face Datasets library and browse the HTML preview files to examine sample data.
数据集概述
数据集名称:CoCreator Driving Scene
数据集地址:https://huggingface.co/datasets/NIyueeE/cocreator-driving-scene
数据集简介:这是一个由CoCreator自动构建的驾驶场景因果推理数据集,专注于驾驶场景理解。
数据集规模
- 样本数量:1227个
- 规模类别:1K < n < 10K
数据内容
每个样本包含以下三个字段:
| 字段 | 类型 | 描述 |
|---|---|---|
id |
字符串 | 样本ID,范围为0001–1227 |
video_frames |
PIL.Image列表 | 按时间顺序排列的帧序列,图像尺寸为1600×900 |
causal_text |
字符串 | 因果描述文本 |
任务类别
- 视觉问答(visual-question-answering)
- 图像到文本(image-to-text)
具体任务包括:图像描述(image-captioning)和视觉问答(visual-question-answering)。
标签与领域
- 驾驶(driving)
- 因果推理(causal-reasoning)
- 自动驾驶(autonomous-driving)
- 场景理解(scene-understanding)
数据来源与标注
- 数据来源:原始数据(original)
- 标注方式:机器生成(machine-generated)
语言
- 单一语言:英语(en)
许可证
- MIT许可证(mit)
数据加载示例
python from datasets import load_dataset
ds = load_dataset("NIyueeE/cocreator-driving-scene")
数据预览
可通过浏览 review.html 文件查看数据预览。




