Egocentric-10K - Build AI's open source first-person perspective robotics dataset

What is Egocentric-10K

Egocentric-10K is a large-scale first-person view (egocentric) factory operation video dataset open-sourced by the build.ai team. The dataset contains 10,000 hours of video, totaling 1.08 billion frames, and involves 2,138 workers, each contributing an average of about 4.68 hours of video content. The dataset consists of 192,900 video clips, each with a median length of 180 seconds and a storage size of 16.4 TB in H.265/MP4 format with 1080p resolution and 30fps frame rate, featuring high-density operating scenes and high-frequency hand visibility, which is a significant improvement compared to previous field datasets.

Features of the Egocentric-10K

• Large-scale dataThe first-person view of the researchers was the first to be used in the study: 10,000 hours of video, totaling 1.08 billion frames, provided researchers with a massive amount of first-person perspective data.
• Real factory environment: The data are completely collected from real factory scenarios, which are highly practical and realistic for industry-related research.
• High-density operation: Frequent and high visibility of hand manipulation in videos makes the manipulation scenes more intensive compared to traditional datasets, which is suitable for action recognition and task learning.
• Diverse Worker Engagement: 2,138 workers were involved, each contributing an average of 4.68 hours of video, with data from a wide range of sources covering a wide range of operating styles and habits.
• Efficient storage and formattingThe H.265/MP4 format, with a resolution of 1080p and a frame rate of 30fps, ensures video quality while optimizing storage space.
• Easy to access and use: Data is organized in WebDataset format for fast loading and processing, suitable for large-scale machine learning and data analysis.
• Multi-disciplinary applications: It is applicable to a wide range of fields such as robot learning, industrial vision, motion recognition, etc., and provides strong support for the development and research of related technologies.

Core Benefits of the Egocentric-10K

• Real Scenario Data: Completely collected in a real factory environment to ensure the high authenticity and practicality of the data, suitable for industrial scenario research.
• Massive data volumeThe research is a rich source of material: 10,000 hours of video, totaling 1.08 billion frames, are included.
• High-density operation: Dense operating scenes and high hand visibility in videos are suitable for action recognition and task learning.
• Diverse data sources: 2,138 workers were involved, and the data covered a wide range of operating styles and habits, making it broadly representative.
• Efficient data formats: H.265/MP4 format to optimize storage and transmission efficiency while maintaining high quality.
• easy-to-use: Organized in WebDataset format for fast loading and processing, suitable for large-scale machine learning.

What is the official website for Egocentric-10K?

• HuggingFace Model Library:: https://huggingface.co/datasets/builddotai/Egocentric-10K

People for Egocentric-10K

• Robotics researchers: Can be used to train and optimize robots to operate in industrial environments, helping them to better understand and perform tasks.
• Computer vision specialists: Provides rich first-person view data for developing and testing industrial vision systems, enhancing the system's ability to recognize and analyze in complex environments.
• Artificial Intelligence Developers: Provide large-scale training data for machine learning and deep learning models to support algorithm development and optimization.
• Industrial Automation Engineer: Contribute to the research and development of more efficient automation solutions to improve the efficiency and quality of industrial production.
• Academic researchers: To provide high-quality data support for academic research in related fields and to promote the development of theoretical and applied research.
• Industrial Data Analyst: It can be used to analyze worker operating behavior, optimize workflow, and improve productivity and safety.