#Japan #Tokyo #AI Development #Qlean Dataset #Visual Bank

Introduction

Visual Bank, a startup based in Shibuya, Tokyo, has recently announced the addition of a new walking video dataset to its Qlean Dataset offerings. This dataset includes videos showcasing people of various ages, genders, and nationalities walking in different settings. With the growing need for diverse training data in artificial intelligence (AI) development, this addition aims to meet the requirements of both research and commercial applications. The dataset is part of Visual Bank's initiative to provide flexible and readily usable AI training resources.

About the Qlean Dataset

The Qlean Dataset consists of original data tailored for commercial use. Its unique configuration allows for a flexible combination of data materials that can be adapted based on purpose, accuracy, and deadlines. Some data are pre-annotated while others are not, enabling modifications to meet specific requirements. Through partnerships with various organizations, such as the Chiba Lotte Marines and Toyo Keizai Inc., as well as extensive domestic and international networks, Visual Bank continues to enhance its lineup of available data, greatly reducing the burden of data collection and organization in AI development environments.

Overview of the Walking Video Dataset

The new walking video dataset includes four distinct versions, each recorded in different environments:

1. Indoor Multinational Version
- Content: Videos of foreign individuals walking, showcasing various decorative elements and capturing right and left turns.
- Sample Attributes: 116 people from diverse backgrounds including Africa, Sri Lanka, India, the Middle East, Europe, the United States, Australia, China, and Southeast Asia.
- File Count: 696 (mp4 format)
- Details: View Sample

2. Indoor Japanese Version
- Content: Videos of Japanese individuals walking with different accessories, including glasses.
- Sample Attributes: 470 individuals ranging from children under ten to seniors over fifty, including 14 pairs of identical twins.
- File Count: 2,820 (mp4 format)
- Details: View Sample

3. Indoor Green Background Japanese Version
- Content: Videos of Japanese individuals captured against a green backdrop, facilitating ease of background removal.
- Sample Attributes: 200 subjects aged from ten to over seventy.
- File Count: 1,200 (mp4 format)
- Details: View Sample

4. Indoor Meeting Room Japanese Version
- Content: Japanese individuals walking in various ways with different accessory combinations.
- Sample Attributes: 70 people of varying ages.
- File Count: 1,050 (mp4 format)
- Details: View Sample

Use Cases of the Walking Video Dataset

The walking video dataset serves multiple purposes, including:

• - Human Flow Analysis and Path Optimization: By capturing walking patterns in environments emulating commercial facilities and transit systems, this dataset aids AI in predicting foot traffic and congestion analysis.

• - Non-Face-Based Person Identification AI Training: This dataset facilitates the development of identification systems that rely on gait and physical characteristics rather than facial recognition, which is crucial in environments like train stations and airports.

• - Assessment of Gait Recognition AI: The diversity in subjects captures cultural and physical differences in walking, which can be used to validate and refine gait recognition technologies.

• - Anomaly Detection Model Training: Typical walking patterns across various ages provide an excellent foundation for developing systems to detect abnormal movements such as falls or instability, which is essential for healthcare environments.

• - Behavior Classification and Attribute Estimation AI Training: The dataset includes distinct scenarios and various walking parameters, ideal for training AIs to classify behavior or estimate attributes of individuals based on their manner of walking.

• - Synthetic Simulation Support Using Green Backdrop Data: The videos shot against a green background can aid in producing realistic simulations in virtual environments, enhancing applications in augmented and virtual reality.

• - Twins and Similar Individuals Gait Distinction: Including data from identical twins provides unique opportunities to assess whether gait can be used as an identifying marker for very similar-looking individuals in security settings.

Upcoming Seminar on Data Utilization

Visual Bank is also hosting a seminar on August 28, 2025, focusing on data-centric practices to enhance accuracy and safety in VLM (Visual Language Model) development. It will cover aspects of high-quality caption creation, intellectual property risks, and privacy concerns which are vital for anyone involved in the field of AI development.

Visit the seminar registration page here: Seminar Registration

Conclusion

With the introduction of this varied walking video dataset, Visual Bank is not just expanding its offerings but is also paving the way for groundbreaking advancements in AI development, addressing the critical need for diverse and comprehensive training data. This initiative emphasizes the intersection of AI with real-world applications, aiming to enhance performance across multiple sectors.

Company Background

Visual Bank is committed to building next-generation data infrastructures that maximize AI development capabilities. Operating with the mission of unlocking the potential of all data, Visual Bank provides services such as the Qlean Dataset through its subsidiary Amana Images. Furthermore, Visual Bank's inclusion in Japan's national research and development program