Recently, I've been showing my game footage to AI. I'm playing around with feeding it game videos to automatically color-code players and map them onto a court layout. Lol. Even though I'm not an engineer, I was able to set it up and get it running by myself in about half a day, so I thought I'd share!
This article contains a lot of technical talk, so if you want to try the challenge, read at least halfway! If you're interested, please read to the end!
Before I introduce it, let me make one announcement! Although I am an active basketball player, I also run Fantrance, a live streaming platform specialized for sports! Athletes talk about their own performance after games, so it's full of stories you can only hear here! Please check it out!
Now, this article is about how I used existing technology and AI to create something that works, even without being an engineer!
What I used this time is a basketball analysis tool within the image recognition AI development platform published by Roboflow, a US AI company.
You might have seen the video above, but I put an ENEOS Sunflowers match into this analysis tool, and the AI automatically labeled the players in red uniforms and yellow (green) uniforms as "ENEOS Sunflowers" and kept tracking each player.
I didn't need to teach it beforehand that "red is here" or "yellow is ENEOS"; the AI learned the uniform colors for that match on its own.
Since entering the roster was a hassle, I just set it from 0 to 99. The referee ended up being identified too, but since I can fix that later, I left it for now.
This is as far as it goes for now, but since it can also detect the ball and whether a shot was made or not, my goal is to be able to extract all kinds of data, such as how many meters someone ran, where they spent the most time on the court, and where they took shots.
I also tried mapping the players' movements!
Furthermore, I participated and presented this initiative as the only athlete at an AI study group where entrepreneurs gathered recently.
In terms of accuracy and what can be done, there's still a long way to go, but... for now, what I want to say is that it's incredibly interesting. Lol. I'm well aware of the concerns about security and other things being said.
But recent AI is so much fun.
I probably have too many things I want to create, so I try this and that, but seeing the quality of AI output getting higher and higher makes me want to give shape to the "interesting ideas" I come up with one after another. For me, AI is the best toy.
By the way, when I do things like this, my engineers have been looking at me nervously lately. Lol.
Because I'm constantly running AI and creating all sorts of projects and talking about them endlessly, they seem to think, "Evelyn might accidentally do something that affects the production environment." Lol.
No matter how many times I explain that I keep my PCs completely separate, they haven't let their guard down. Lol.
[How Athletes Should Engage with AI]
I'm not an engineer, and I think I'm doing something very clumsy from the perspective of a real developer. I know people might say, "You're just clicking around in a browser" or "That technology has existed for a long time."
However, for me just a little while ago, implementing this much was quite a high hurdle and required time.
Even so, catching up with cutting-edge AI and touching it on a daily basis makes me think that there is value in "touching technology even if you don't quite understand it."
When you talk to AI about what's possible on a daily basis, your own competition footage gradually starts to look like "data" rather than just a record. You start to think, "Maybe there's a correlation between this stat and this play!" Even if that's already an established data analysis method, you'll be able to see your sport from a new perspective.
And it's an athlete's privilege to have more body data than anyone else. There's a possibility that things that were once just intuition can lead to a clearer understanding of yourself.
"Shouldn't athletes be the ones touching AI?" I truly thought so this time, so here is my proposal!
I'd like to show "5 Attitudes Necessary for Athletes in the AI Era."
- First, be able to think about experimenting with something using AI
- Think about how to solve things when they don't work
- Imagine how new technology connects with the fields you love
- How much can you increase the number of unrelated fields you like?
- Can you jump into knowledge and networks in fields you know nothing about?
I believe these five points are very important for athletes to be unique and to live in this era!
"I didn't think Evelyn was doing AI to this extent." I've been told this very often lately. I think I'll be told this even more from now on.
Athlete x Entrepreneur x AI Implementation. I wonder how many people in Japan are doing all of this with their own hands. If there aren't any, please let me call myself a "Tech Athlete." Lol.
For those interested in the technology, I'll explain a bit more in detail below. If you're not interested, you can skip it. (As far as I can write as a non-engineer.)
And! Since I'm touching AI almost every day like this, I'd love to talk about various things! Are there any athletes or sports-related people who are also interested in technology?
Conversely, if there's anything like "Can't we do this with tech x sports?" please feel free to consult with me! 🙋🏽♂️
And to engineers who are doing AI analysis of video like this, or those familiar with motion analysis! If you'd like, please let me hear your thoughts.
That's all for Tech Athlete Evelyn Mawuli's AI struggle report!
--- For those interested, please see below! ---
Some of the AI tools I use every day:
- Claude Code (for text planning and as an advisor)
- Codex (code generation)
- ChatGPT (for quick consultations)
- OpenClaw (an AI agent I recently made. Named Shaq)
What I used this time is a basketball analysis tool within the image recognition AI development platform published by Roboflow, a US AI company.
In addition:
- RF-DETR (AI that detects players with high precision)
- Meta's SAM2 (AI that tracks each player until the end of the match)
- Google's SigLIP (AI that automatically teams up based on uniform color)
I ran these on Google Colab Pro+ using an NVIDIA L4 GPU.
However, to be honest, it didn't go smoothly at all. You'd think that published AI notebooks would just work if you press a button...
I was taught that after about six months, they're usually broken. Lol.
What I stumbled on was:
- Infinite loop due to version conflict of Pillow (image processing library) -> Compatibility between image processing tools was bad, and errors looped.
- SAM2 build error -> Initial setup to run the AI model didn't go through well.
- Error due to API change in the jersey number recognition model -> Due to specification changes in external tools, the code suddenly stopped working.
What annoyed me the most was:
- The AI assistant (Gemini) in Colab, which explains errors and fixes code, was quite wrong. Lol.
When I gave screenshots to Codex or Claude Code, they said, "This guy is saying the wrong thing, so consult me before consulting him," and I was fixing it while laughing, thinking that this kind of workplace probably exists. Lol.
But what I actually do isn't that difficult; it's just a repetition of executing cells and asking the AI when an error occurs.
In the end, it detects jersey numbers from the video, cuts out moments, and generates a color-coded video. An engineer could probably do it in 30 minutes.
So, the trick for non-engineer athletes or people like me to achieve something with AI is to just keep asking. Don't be embarrassed to ask the AI. Moreover, if you continue, it becomes a learning experience, so don't think it's a waste and just keep asking until you understand!
Since I'm touching AI almost every day like this, if there are any athletes or sports-related people interested in technology, let's talk!
Conversely, if you have any consultations like "Can't we do this?" please let me know! 🙋🏽♂️
Back Dooor Inc. aims to expand not only within Japan but also overseas in the near future. Wonderful players exist all over the world, and there are the best fans there. Because sports is a universal content, I believe the market is large and there is significance in expanding. To that end, we will focus on hiring engineers to create products, sales positions, and human resources to promote business globally.
The usefulness of perceiving sports as a business and the market size are starting to be recognized in Japan, so now is the best timing to compete.
- I want to make use of my experience in sports.
- I want to give back to the sports industry.
- I am interested in sports business. If you are such a person, please! I would be happy if you could contact Back Dooor Inc.
That's all for Tech Athlete Evelyn Mawuli's AI struggle report!
