We used harBest to create training data for object detection/annotation and we succeeded in accelerating AI development.

 What you'll learn about in this article:

・Satellite images & AI project challenges and solutions
・The importance of quality control in annotation data

We spoke to Mr. Imagawa of ‘Penetrator’ a startup company from JAXA whose vision is to solve real estate issues from space.

First of all, can we ask what your company does?

We are creating, in collaboration with JAXA, a product called ‘WHERE’, which uses AI to locate valuable real estate from space. When building a house or building, you first need to purchase land directly from the owner. Purchasing land is a very important task that is fundamental to the real estate business. However, many real estate companies still rely on the legacy method of personal connections, making it extremely difficult to purchase prime land. That network of connections is private and is not made publicly available. In order to find this kind of information, real estate agents have to search for land and conduct land surveys. Searching for land requires a lot of physical effort, and the land survey itself involves going to the Legal Affairs Bureau multiple times, requesting documents, and other complicated tasks, and it takes about 30 hours to create a sales list of 100 properties. The research that I conducted at JAXA provided a hint as to how to solve these issues. It involved using AI to analyze craters in lunar satellite data. I started to wonder if the technology could be applied to analyze data here on Earth instead of the moon, and real estate instead of craters. That was the beginning of ‘WHERE’.

What kind of product is it?

As a solution it’s very simple. It uses AI to pick out vacant lots, fields, possible vacant houses etc. from satellite data. In addition, since it is linked to the owner data of the Legal Affairs Bureau, the information is available immediately and the owner can be approached straight away. We are looking for Object Detection annotations for this. Currently, the logic is that AI can be used to find a wide variety of things and, from there, another AI can be used to recommend the land that the user is looking for.

Can we ask about your background?

During my time at university, I was involved in some research that combined AI and security. I was very interested in AI, so I wanted to apply it to other fields. When I was looking for laboratories, I decided to join Professor Tanaka’s laboratory because I would also be able to carry out research in relation to space. Even though Professor Tanaka was working at the University of Tokyo, he was also a professor at the JAXA Institute of Space and Astronautical Science, where he conducted research. There was also a JAXA-affiliated graduate school system where I did research together with Professor Tanaka.

How did you meet Mr. Akutsu, who also works at ‘Penetrator’?

Mr. Akutsu had been running a real estate company for a long time, and was frustrated by the process of purchasing land. He had also joined JAXA, apparently in order to see the world from space and gain a new perspective. He was working in Professor Tanaka’s laboratory when I met him. He asked me if I could apply the moon research to land purchasing, so I said, “I’ll give it a try”. Within about a week, I had created a prototype and we both thought “We can do this!”. That’s how we got to this point.

Can you tell us about your current position?

I work as a development ‘scrum master’ and also manage the engineers. I also manage the KPI numbers company-wide and I collaborate with external parties to strengthen tech branding.

How did you find out about harBest?

We receive funding from the University of Tokyo IPC. As we were thinking about outsourcing our annotation work, someone there mentioned that they knew a company who could help, and so we contacted APTO via your website.

Did you consider other annotation companies?

We had considered either hiring someone permanently to handle annotation or using freelancers and crowd workers. We had also consulted some annotation companies at events.

What made you decide to use us?

The main deciding factor was Mr. Takashina (APTO CEO) and his enthusiasm. But also, the person in charge of customer success was managing the project very carefully and we were very satisfied with the quality.

Thank you very much! Could you tell us about your plans for the future?

We are aiming to make ‘WHERE’ a unicorn company. We want to fundamentally change the process of buying land. As I said, I want to change the real estate market across the globe within the next ten years. I’d like to create a product that can be used by non-real estate companies and move into the BtoBtoC field. In the future, we would like to explore not only Earth but also real estate on Mars. In order to achieve our vision, we are looking beyond the Japanese market and into overseas real estate markets

Are there any other developments on the horizon?

We have recently obtained owner information from across the United States, so that is quite a big deal. We will be exhibiting at the CES 2024 exhibition in January, so we would like to carry out some market research there and continue with POC.

Thank you very much for talking to us! We will keep improving harBest and we look forward to continue working with you!