Case Studies

Tell us about your business.

SkymatiX is a company specializing in remote sensing and GIS. Remote sensing involves obtaining information from a distance, such as observing the ground from the sky and performing various analyses based on the images.

Additionally, we work with geographic information systems (GIS), enhancing various information on maps to add value naturally. While most companies specialize in one area, SkymatiX differentiates itself by excelling in both.

What kind of data creation were you doing before using our services?

We have used services from several similar annotation companies. When Japanese workers perform the annotation tasks, the costs can be high, so we often outsourced to overseas vendors. For small quantities, we sometimes handled it in-house. We made decisions based on the volume and cost-effectiveness.

In regards to workload, how do you decide whether to handle it in-house or outsource?

It depends on the volume that feels overwhelming to handle alone. Some tasks are heavier than others. For example, the tasks we requested from your company were relatively heavy, making it challenging to annotate even a single satellite image. In such cases, we decided to outsource even just five satellite images. On the other hand, for simple tasks like marking objects, we often handled up to 100 items in-house.

Deadlines also play a role. If the deadline is tight, communication can take time, so we sometimes handle small tasks in-house. The main criterion is whether it’s too burdensome to do it ourselves.

Outsourcing involves creating manuals, holding meetings, and checking deliverables, so we compare which option is more challenging before deciding to outsource.

You mentioned using overseas annotation vendors; what challenges did you face in doing so?

There are no particular issues. In our case, we often deal with images of vegetables and the ground, not personal information like faces, making it easier to use. However, when dealing with photos inside a factory under an NDA, it was difficult or impossible to send them overseas. In such cases, we decided to handle it ourselves.

In this regard, YAZAKI’s services provide a sense of security and compliance management, which we appreciate.

Were there any challenges or bottlenecks when requesting services from other annotation vendors?

Yes, when outsourcing, we ultimately use the delivered products, so the quality is crucial. Sometimes, we need to spend time checking and requesting revisions.

What types of objects did you find required the most revisions?

This includes satellite images and aerial photographs. Some tasks are very simple. For example, in cases where you just need to circle cabbages, the instructions are easy for the requester to give and clear for the recipient to understand.

The task we requested from YAZAKI this time included determining the boundaries of rivers, which was quite complex. Although we should have created a manual beforehand, there were many parts that we couldn’t understand until we tried. In that sense, it might have been challenging.

The most challenging task we anticipated was annotating trees. From above, grass and trees look the same color, so the difference lies in the shadows created by the height of the trees. Therefore, we asked them to find and annotate those shadows, which was quite difficult.

How important is the initial requirement definition?

Yes, before conducting the overall annotation, YAZAKI asked us, “How about this?” which positively impacted the later stages of the process.

Some companies offer services characterized by being “fast” or “cheap,” but when you check, they rush through the work. For example, if you request 10 items, upon delivery, you might find many corrections needed, leading to the conclusion that “it would be faster to do it myself.”

Annotation companies offer both “cheap and fast” services and those that are a bit more expensive but save time in the long run. Deciding which is more beneficial can be challenging. When we need careful work, we consider asking YAZAKI.

What made you decide to use our service?

Yes. Initially, your company approached us, saying, “We are confident in our quality,” so we decided to give it a try. It was also the timing of a large project, and we were unsure if our existing annotation company could handle it alone, so we wanted to try different companies.

What did you find beneficial about using our services?

The biggest advantage was that you created good annotation data without much effort on our part. I think it’s because you spent time on requirement definition at the beginning. Once that was done properly, we could tell you to do the rest the same way, saving us effort. The delivered data was top quality compared to other companies.

Is it important to have engineers handle some tasks for accuracy, and how do you balance this?

Yes. When explaining the desired annotation to others, creating a manual and explaining it can be time-consuming. For about 100 images, it might be faster to do it myself. However, for a volume of 10,000 images, it becomes challenging, so we might decide to outsource even if it requires effort to explain. It’s a balance issue.

Documentation skills for creating manuals are also important (※2). If the manual is unclear, the person asked to do the work might get confused, leading to more inquiries and incorrect deliverables. The effort spent here can be a concern.

What do you think about our proposal skills and requirement definition before starting the work?

We were very satisfied. When we heard that YAZAKI was doing annotations, we hesitated because the sample images were different from satellite images.

When we asked, “Can you handle this kind of work?” you said, “Yes, we can,” so we decided to give it a try, especially since we had a large volume. The proposals based on our development requirements were also good.

What are your expectations from us moving forward?

Annotation vendors fall into two categories: those that accept standard tasks and those that take on complex, unique tasks. YAZAKI is the latter, a company that is willing to consult and has good responsiveness. We expect higher technical skills in the future. We may request high-difficulty annotations, such as those with challenging contour identification or high confidentiality, so we look forward to working with you.

* 1: Annotation is the process of adding meaning (tagging) to data that AI wants to learn.
*2: We also support the creation of operating procedures.