BAAI-DataCube_DIY视频数据集

## 数据集简介

* 智源数据魔方（ https://datacube.baai.ac.cn ）系列数据集由智源数据魔方产品根据不同场景和需求自动构建，欢迎加入用户沟通群


* 涵盖7大方向，40个高质量视频数据，详情如下：

| 系列 | 子方向 | 数据集中文名称 | 数据集英文名称 | 应用场景 |
| ------- | ------- | ----------- | -------------------------------- | --------------------------------------------------- |
| 人物动作与交互 | 手部动作 | 手部动作视频数据集 | Gesture-Recognition | 用于手势识别模型、人机交互模型，VR/AR 手势控制、智能家居指令识别等场景 |
| | | 手工/DIY视频数据集 | DIY | 用于操作步骤识别、手部动作跟踪等场景，适合训练机器人 学习装配/制作任务，也可用于动作分解和模仿学习。 |
| | | 烹饪视频数据集 | Cooking | 适合动作分解学习（切菜、翻炒等），任务规划（烹饪步骤预测）等场景 |
| | 身体动作 | 健身运动视频数据集 | Fitness-Exercise | 用于动作识别与纠错模型，支持健身指导、姿态矫正、智能健身教练等场景 |
| | | 室外运动视频数据集 | Outdoor-Sports | 适合人体关键点识别、动作分析等场景，支持体育赛事分析、智能裁判和运动辅助机器人等应用 |
| | | 儿童活动视频数据集 | Children’s-Activities | 用于行为检测、儿童安全监控等场景，适合教育机器人、陪伴机器人等应用 |
| | 表情与情绪 | 情绪视频数据集 | Facial-Expression | 用于训练情绪识别模型，支持人机交互情感计算、客服机器人、心理健康监测等场景 |
| | | 表情视频数据集 | Fine-grained-Facial Expression | 细粒度表情识别数据集，支持微表情分析、多模态对话系统等应用，提升 LLM 在具身交互中的情感理解能力 |
| | 社会交互 | 双人互动视频数据集 | Two-person-Interaction | 训练社交互动识别等模型，支持对话场景检测、机器人社交技能模拟等场景 |
| | | 人群聚集视频数据集 | Crowd | 用于人群检测、密度估计、城市安防、公共场所监控中的异常检测等场景 |
| | | 校园视频数据集 | Campus | 训练群体交互识别模型，支持校园安全、课堂场景分析、教育 AI 系统建设等应用 |
| 第一人称与导航 | 驾驶与第一人称 | 第一人称驾驶视频数据集 | First-person-Driving | 自动驾驶和第一人称导航任务，支持 驾驶行为预测、路径规划等应用 |
| | 导航与镜头移动 | 室内场景视频数据集 | Indoor-Scene-Dataset | 支持训练室内导航等模型，支持服务机器人在 室内环境中导航、避障等任务 |
| | | 公路场景视频数据集 | Road-trip-Dataset | 用于道路场景感知、自动驾驶场景模拟等任务和场景 |
| 表演与舞台 | 舞蹈与表演 | 舞蹈表演视频数据集 | Dance-Performance Dataset | 用于人体动作生成、舞蹈动作模仿等场景 |
| | | 歌曲表演视频数据集 | Song-Performance Dataset | 用于歌唱动作识别、舞台表演动作预测等应用 |
| | 舞台/演讲 | 演讲视频数据集 | Speech-Dataset | 训练姿态-语音多模态模型，支持 演讲者手势分析、虚拟主持人等场景 |
| | | 表演视频数据集 | Performing-Arts-Dataset | 支持虚拟表演生成、数字人训练、舞台动作理解等场景 |
| | | 新闻播报视频数据集 | Breaking-News-Dataset | 支持虚拟新闻主播生成等应用 |
| 热点与节日 | 情人节 | 情人节日视频数据集 | Valentine’s-Day-Dataset | 训适合广告推荐、节日事件检测、情感交互等应用 |
| 安全与安防 | 火灾 | 火灾视频数据集 | Fire-Incident-Dataset | 训练火灾检测等模型，支持智能监控、机器人应急响应 |
| | 洪水 | 洪水视频数据集 | Flood-Dataset | 训练自然灾害识别等模型，支持防灾预警等应用 |
| | 加油站 | 加油站视频数据集 | Gas-Station-Dataset | 训练特定场景监控等安全模型，支持安全生产和异常行为检测等应用 |
| | 劳动安全 | 工人穿戴视频数据集 | Worker-Wearing-Detection-Dataset | 训练安全检测模型，支持工业安全与合规监控等应用 |
| 生活与日常 | 家庭生活 | 家庭生活视频数据集 | Living-Room-Dataset | 训练 日常行为识别等模型，适合家庭服务机器人、智能安防等应用 |
| | 儿童教育 | 儿童学习视频数据集 | Children-Classroom Dataset | 训练 教育场景行为识别等模型，支持课堂互动检测、教育机器人等应用 |
| 动物与自然 | 动物 | 鸟视频数据集 | Bird-Dataset | 训练动物识别等模型 |
| | | 猫视频数据集 | Cat-Dataset | 支持动物识别与分类、宠物行为分析、视频监控中小动物检测等应用 |
| | | 狗视频数据集 | Dog-Dataset | 犬类动作识别、支持导盲犬/工作犬行为建模、宠物表情识别、多模态视频问答等应用 |
| | | 鱼视频数据集 | Fishes-Dataset | 支持水下检测与识别、海洋生物监测等应用 |
| | | 昆虫视频数据集 | Insect-Dataset | 支持生态环境监测、害虫检测与农业智能化、微小目标检测与跟踪研究等应用 |
| | 自然 | 雨林视频数据集 | Rainforest-Dataset | 支持复杂场景识别、环境变化监测、生态保护研究等应用 |
| | | 沙漠视频数据集 | Desert-Dataset | 支持极端环境识别、自动驾驶在稀疏地貌中的视觉感知训练、气候变化研究 |
| | | 山川视频数据集 | Mountain-Dataset | 户外场景识别、无人机导航、AR/VR 场景生成、地貌分类 |
| | | 河流视频数据集 | River -Dataset | 支持水域检测与监控、水文环境建模、洪涝灾害监测等应用 |
| | | 海洋视频数据集 | Ocean-Dataset | 支持海洋场景分割、水下机器人感知、海洋生物识别等模型训练和应用 |
| | | 草原视频数据集 | Grassland-Dataset | 应用于大面积生态环境识别、无人机遥感任务、野生动物栖息地分析等模型 |
| | | 天空视频数据集 | Sky-Dataset | 应用于天气识别、航拍与卫星视觉、自动驾驶视觉背景建模等场景 |
| | | 森林视频数据集 | Forest-Dataset | 森林防火监控、无人机林业巡检、复杂植被环境建模等场景 |
| | | 花视频数据集 | Flower-Dataset | 植物识别与分类、园艺/农业智能化、细粒度图像识别、多模态美学生成等场景 |
| | | 雨视频数据集 | Rain-Dataset | 天气感知与预测、视觉去雨任务训练、自动驾驶在恶劣天气下的测试等场景 |

## 数据魔方产品介绍

* 智源研究院研发的数据魔方（DataCube）可实现从“数据集Level”到“数据样本Level”的精准检索，满足用户个性化的数据需求，只需自然语言输入数据需求，平台便能快速构建专属的个性化数据集，彻底打破以往操作流程的桎梏。

* 在核心技术架构层面，数据魔方依托智源数据平台支持的100+数据处理算子，实现 PB 级数据的自动化处理Pipeline；深度融合 CLIP Understanding 引擎，对多模态样本从本体、行为、视角与风格等多维细进行细粒度语义解析，逐条构建精细数据画像；辅以 Hybrid Retrieval 体系，实现跨模态特征的毫秒级精准召回；最终通过 Data Evaluation 评估数据集构建效果，实现个性化高质量数据集生成。数据魔方具有以下特点：

*  响应速度“快”，自然语言输入需求后，最快能秒级反馈并生成数据集

* 成本“省”，免去繁琐的数据筛选过滤过程，极大降低了人力与时间成本

* 数据量“多”，目前已汇聚5000W+数据样本基础，且数量仍在持续增长

* 数据质量“好”，内置深度语义理解算法，确保检索到的数据相关性极高

* 数据魔方使用演示：

[数据魔方演示-DIY-8月31日.mov](https://resources.ks3-cn-beijing.ksyuncs.com/projects/DataCube/demo.mov)

## 许可与使用

* 本数据集仅限学术研究与非商业用途

* 使用即视为同意智源研究院数据魔方产品相关协议和使用条款

* 意见反馈与业务合作，欢迎联系：<[e-mail]>&#x20;

## Dataset Introduction
* The DataCube series datasets (https://datacube.baai.ac.cn) are automatically constructed by the BAAI DataCube product based on different scenarios and requirements. Welcome to join the user communication group

* Covers 7 major categories and 40 high-quality video datasets, details are as follows:

| Series | Sub-direction | Chinese Dataset Name | English Dataset Name | Application Scenarios |
| --------------------- | ------------------- | -------------------------- | --------------------------------------------- | ------------------------------------------------------------------------------------- |
| Human Motion & Interaction | Hand Motions | Hand Gesture Video Dataset | Gesture-Recognition | For scenarios such as gesture recognition model training, human-computer interaction model training, VR/AR gesture control, smart home instruction recognition, etc. |
| | | Handcraft/DIY Video Dataset | DIY | For operation step recognition, hand motion tracking and other scenarios, suitable for training robots to learn assembly/production tasks, and also for action decomposition and imitation learning. |
| | | Cooking Video Dataset | Cooking | Suitable for action decomposition learning (chopping vegetables, stir-frying, etc.), task planning (cooking step prediction) and other scenarios |
| | Body Motions | Fitness Exercise Video Dataset | Fitness-Exercise | For motion recognition and error correction models, supporting scenarios such as fitness guidance, posture correction, intelligent fitness coaches, etc. |
| | | Outdoor Sports Video Dataset | Outdoor-Sports | Suitable for human keypoint recognition, motion analysis and other scenarios, supporting applications such as sports event analysis, intelligent referees and sports-assisted robots |
| | | Children's Activities Video Dataset | Children’s-Activities | For behavior detection, children's safety monitoring and other scenarios, suitable for applications such as educational robots, companion robots, etc. |
| | Facial Expression & Emotion | Emotion Video Dataset | Facial-Expression | For training emotion recognition models, supporting scenarios such as human-computer interaction affective computing, customer service robots, mental health monitoring, etc. |
| | | Fine-grained Facial Expression Video Dataset | Fine-grained-Facial Expression | Fine-grained facial expression recognition dataset, supporting applications such as micro-expression analysis, multimodal dialogue systems, etc., to improve the emotional understanding ability of LLMs in embodied interaction |
| | Social Interaction | Two-person Interaction Video Dataset | Two-person-Interaction | For training social interaction recognition and other models, supporting scenarios such as dialogue scene detection, robot social skill simulation, etc. |
| | | Crowd Gathering Video Dataset | Crowd | For crowd detection, density estimation, urban security, anomaly detection in public place monitoring and other scenarios |
| | | Campus Video Dataset | Campus | For training group interaction recognition models, supporting applications such as campus security, classroom scene analysis, educational AI system construction, etc. |
| First-person Vision & Navigation | Driving & First-person View | First-person Driving Video Dataset | First-person-Driving | For autonomous driving and first-person navigation tasks, supporting applications such as driving behavior prediction, path planning, etc. |
| | Navigation & Camera Movement | Indoor Scene Video Dataset | Indoor-Scene-Dataset | Supporting training indoor navigation and other models, supporting tasks such as service robots navigating and avoiding obstacles in indoor environments |
| | | Road Trip Video Dataset | Road-trip-Dataset | For road scene perception, autonomous driving scene simulation and other tasks and scenarios |
| Performing Arts & Stage | Dance & Performance | Dance Performance Video Dataset | Dance-Performance Dataset | For scenarios such as human motion generation, dance motion imitation, etc. |
| | | Song Performance Video Dataset | Song-Performance Dataset | For applications such as singing motion recognition, stage performance motion prediction, etc. |
| | Stage/Speech | Speech Video Dataset | Speech-Dataset | For training posture-speech multimodal models, supporting scenarios such as speaker gesture analysis, virtual hosts, etc. |
| | | Performing Arts Video Dataset | Performing-Arts-Dataset | Supporting scenarios such as virtual performance generation, digital human training, stage motion understanding, etc. |
| | | Breaking News Video Dataset | Breaking-News-Dataset | Supporting applications such as virtual news anchor generation |
| Hot Events & Festivals | Valentine's Day | Valentine's Day Video Dataset | Valentine’s-Day-Dataset | Suitable for applications such as advertisement recommendation, festival event detection, emotional interaction, etc. |
| Safety & Security | Fire | Fire Incident Video Dataset | Fire-Incident-Dataset | For training fire detection and other models, supporting intelligent monitoring, robot emergency response, etc. |
| | Flood | Flood Video Dataset | Flood-Dataset | For training natural disaster recognition and other models, supporting applications such as disaster prevention and early warning |
| | Gas Station | Gas Station Video Dataset | Gas-Station-Dataset | For training specific scene monitoring and other security models, supporting applications such as safe production and abnormal behavior detection |
| | Labor Safety | Worker Wearing Detection Video Dataset | Worker-Wearing-Detection-Dataset | For training safety detection models, supporting applications such as industrial safety and compliance monitoring |
| Daily Life | Family Life | Family Life Video Dataset | Living-Room-Dataset | For training daily behavior recognition and other models, suitable for applications such as home service robots, intelligent security, etc. |
| | Children's Education | Children's Learning Video Dataset | Children-Classroom Dataset | For training educational scene behavior recognition and other models, supporting applications such as classroom interaction detection, educational robots, etc. |
| Animals & Nature | Animals | Bird Video Dataset | Bird-Dataset | For training animal recognition and other models |
| | | Cat Video Dataset | Cat-Dataset | Supporting applications such as animal recognition and classification, pet behavior analysis, small animal detection in video monitoring, etc. |
| | | Dog Video Dataset | Dog-Dataset | Canine motion recognition, supporting applications such as guide dog/working dog behavior modeling, pet expression recognition, multimodal video QA, etc. |
| | | Fish Video Dataset | Fishes-Dataset | Supporting applications such as underwater detection and recognition, marine organism monitoring, etc. |
| | | Insect Video Dataset | Insect-Dataset | Supporting applications such as ecological environment monitoring, pest detection and agricultural intelligence, tiny target detection and tracking research, etc. |
| | Nature | Rainforest Video Dataset | Rainforest-Dataset | Supporting applications such as complex scene recognition, environmental change monitoring, ecological protection research, etc. |
| | | Desert Video Dataset | Desert-Dataset | Supporting extreme environment recognition, visual perception training of autonomous driving in sparse terrain, climate change research |
| | | Mountain Video Dataset | Mountain-Dataset | Outdoor scene recognition, drone navigation, AR/VR scene generation, terrain classification |
| | | River Video Dataset | River -Dataset | Supporting applications such as water area detection and monitoring, hydrological environment modeling, flood disaster monitoring, etc. |
| | | Ocean Video Dataset | Ocean-Dataset | Supporting model training and applications such as ocean scene segmentation, underwater robot perception, marine organism recognition, etc. |
| | | Grassland Video Dataset | Grassland-Dataset | Applied to models such as large-area ecological environment recognition, drone remote sensing tasks, wildlife habitat analysis, etc. |
| | | Sky Video Dataset | Sky-Dataset | Applied to scenarios such as weather recognition, aerial and satellite vision, autonomous driving vision background modeling, etc. |
| | | Forest Video Dataset | Forest-Dataset | Scenarios such as forest fire prevention monitoring, drone forest patrol, complex vegetation environment modeling, etc. |
| | | Flower Video Dataset | Flower-Dataset | Scenarios such as plant recognition and classification, horticulture/agricultural intelligence, fine-grained image recognition, multimodal aesthetic generation, etc. |
| | | Rain Video Dataset | Rain-Dataset | Scenarios such as weather perception and prediction, visual deraining task training, autonomous driving testing in severe weather, etc. |

## DataCube Product Introduction
* The DataCube developed by Beijing Academy of Artificial Intelligence (BAAI) enables precise retrieval from "dataset level" to "data sample level", meeting users' personalized data needs. Users only need to input data requirements in natural language, and the platform can quickly build exclusive personalized datasets, completely breaking the shackles of previous operation processes.

* In terms of core technical architecture, DataCube relies on more than 100 data processing operators supported by the BAAI data platform to build an automated processing pipeline for PB-level data; it deeply integrates the CLIP Understanding engine to perform fine-grained semantic parsing of multimodal samples from multiple dimensions such as ontology, behavior, perspective and style, and build detailed data profiles one by one; supplemented by the Hybrid Retrieval system to achieve millisecond-level precise recall of cross-modal features; finally, evaluate the dataset construction effect through Data Evaluation to generate personalized high-quality datasets. DataCube has the following characteristics:
* Fast response speed: After inputting requirements in natural language, it can feedback and generate datasets in seconds at the fastest.
* Low cost: Eliminates tedious data screening and filtering processes, greatly reducing labor and time costs.
* Large data scale: Currently, it has accumulated a base of more than 50 million data samples, and the number is still growing continuously.
* High data quality: Built-in deep semantic understanding algorithm ensures that the retrieved data has extremely high relevance.

* DataCube Usage Demo:
[DataCube Demo - DIY - August 31.mov](https://resources.ks3-cn-beijing.ksyuncs.com/projects/DataCube/demo.mov)

## License and Usage
* This dataset is only for academic research and non-commercial use.
* By using this dataset, you are deemed to have agreed to the relevant agreements and terms of service of the BAAI DataCube product.
* For feedback and business cooperation, please contact: <[e-mail]>