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

Introduction

Visual Bank Corporation, located in Shibuya, Tokyo, has made significant strides in advancing AI learning data solutions with its latest offering, the Qlean Dataset. Under the leadership of CEO Masanari Nagai, the company has partnered with Amana Images to provide a versatile dataset tailored for research and commercial AI development. Recently, they have introduced a new Product Image Variation Dataset in collaboration with a major domestic e-commerce platform, further broadening their extensive data lineup known as Data Recipes. More details can be found on their site: Qlean Dataset.

What is the Qlean Dataset?

The Qlean Dataset features a selection of original, commercially usable data tailored for AI purposes. Its unique structure allows for flexible combinations of data materials based on specific needs, including both annotated and unannotated data. Additionally, Visual Bank collaborates with partners like the Chiba Lotte Marines and Toyo Keizai, along with a network of domestic and international sources, to enhance this data lineup systematically. This initiative aims to alleviate the burdens of data collection and organization in AI development, thus accelerating the overall development process.

New Product Image Variation Dataset

The recently added Product Image Variation Dataset includes:

• - Data Contents: Detailed images of products across common e-commerce categories such as apparel, miscellaneous goods, and home decor, covering various styles including product shots and usage scenarios.
• - Data Format: JPEG images.
• - Image Count: Over 1 million images.

Sample Images

• - Apparel item photography
• - Interior item usage images

The dataset serves multiple use cases, including:
1. E-commerce Image Generation: By using product shots and varied images, this dataset can automatically generate new compositions, backgrounds, and tones for existing products, addressing the lack of diverse product imagery and enhancing page appeal while reducing operational costs.
2. Advanced Search and Recommendation Algorithms: The dataset’s structure allows for images of the same product to be linked across different backgrounds and angles, facilitating the development of high-precision recommendation systems for similar products and image searches, thereby improving the purchasing journey of e-commerce users.
3. Multi-Modal AI for Product Descriptions: The dataset can be leveraged to train models that generate textual descriptions of product features and usage scenarios from images, supporting automatic generation and summarization of product descriptions that are ideal for LLM collaboration PoCs.
4. Product Category Classification: With diverse product images, this dataset can assist in the classification learning of product attributes such as apparel, miscellaneous items, food, etc. It can be applied to structure e-commerce catalogs and automate internal database organization tasks.
5. Benchmark Development for Product Recognition AI: By including a diverse range of images within a single category, the dataset allows for precision validation under conditions prone to misrecognition, thus aiding in the analysis of existing model performance bottlenecks.
6. Pre-Training for Image Generation Models: Featuring commercially viable e-commerce product images, the dataset allows for safe pre-training or fine-tuning of general image generation models, contributing to improved generation accuracy for specific categories.

Key Features of the Qlean Dataset

• - Adaptable for Research and Commercial Use: All datasets include consent forms from every subject and comply with various privacy policies globally, ensuring safe usage for both research and commercial purposes.
• - Rapid Data Acquisition with ROI Maximization: The unique Data Recipe approach enables a cost-effective way to procure data, minimizing initial investments.
• - Custom Data Solutions: For datasets not listed in the Data Recipes, customization according to specific requirements is also possible, leveraging Qlean Dataset capabilities for tailored datasets.

Academia Support Program

As part of its commitment to the academic community, Visual Bank has launched a free dataset provision program through the Qlean Dataset, targeting universities, research institutions, and nonprofit tech development teams. This initiative provides over 500,000 high-quality, rights-cleared datasets encompassing various formats including images, audio, video, and text to support research efforts. Visual Bank, recognized as a participant in the government’s GENIAC project, is dedicated to solving challenges pertaining to high-quality learning data in research settings. For more information, visit: Academia Support Program.

Company Overview

Visual Bank Corporation is a startup dedicated to constructing and providing next-generation data infrastructure to maximize AI development capabilities. With a mission to unleash the potential of all data, Visual Bank also supports manga artists with AI-assisted tools through its subsidiary Amana Images, which develops AI learning datasets. The company is actively pursuing societal implementation of its innovations as a recognized participant in the GENIAC research and development program.

CEO: Masanari Nagai
Location: 5-3-23 Koji-cho, Chiyoda City, Tokyo, Japan 102-0083, WeWork, Nippon Television Yotsuya Building
Visual Bank Company URL: Visual Bank
Amana Images Company URL: Amana Images