John Com

I make TikToks. I (try to) do soccer analytics. I spend way to much time thinking about #Doop

[John Com @JohnComFootball]

Transfermarkt Market Values are often used in my analysis, but they are imperfect. Some may point to my data being base on TM Market Values as a method of dismissing my argument.

Here I explain why I believe they're a valuable metric in Football Analysis
https://t.co/NvvKCef4qh

[John Com @JohnComFootball]

@skalskip92 Sorry to bug you with maybe a dumb question, but any idea what's going on wrong with "from inference import get_model"?

I've got collab notebooks that suddenly don't work. I even see this notebook has the same issue https://t.co/tm9QXjiHyi

[Marwane Hamdani @Hamdanov98]

📚 New Arrivals! 🚀

After a long wait, these gems have finally arrived!
Can’t wait to dive into them and explore the world of football data analytics and xG insights. 🧠⚽

#FootballAnalytics #DataScience #xG #MustRead #AnalyticsRevolution

[John Com @JohnComFootball]

⚽️ Soccer Vision October! Day 20 Part. 2
Implemented the cleaned ball path into my visualization and it's definitely an improvement! Still wanna take some time to understand what I'm doing here, but I'm glad I was able to "functionize" ball path cleaning!

[John Com @JohnComFootball]

⚽️ Soccer Vision October!
For +2 years I’ve had the idea to use ML computer vision to generate real time football tracking data. The technology to democratize tracking data and make it available to any level from grassroots to professional is here! Someone just has to do it!

I toyed with the idea here & there, but ultimately lacked the necessary ML experience (and quite frankly the time) to be able to complete this project. The idea was basically shelved until SkalskiP and Roboflow blew my mind and dropped a 90 minute tutorial pulling off exactly what I dreamed up.

This has lit a fire under me! For next 30 days I’m diving deep into ML CV, Roboflow tutorials and football tracking data with the goal of making an app that can generate tracking data from any match footage! Follow along on my ML journey as I post a new update here every day!

[SkalskiP @skalskip92]

over 200 hours of work compressed into a 90-minute video

the football AI tutorial is finally out!

link to video: https://t.co/d09LtEL6Sa

↓ key takeaways

[John Com @JohnComFootball]

⚽️ Soccer Vision October! Day 20
Revisited things and was able to use Eric Fenaux's Ball Radar code to improve my ball tracking data. I love how it turned out, but I'm also frustrated, because I feel like I have no idea how it works. The math feels very over my head and a bit magic.

Still gonna incorporate this great work into my project, but I may spend a day or so learning the math/magic behind this so I can understand properly what I'm working with.

For those interested, here's Eric's code: https://t.co/7pZAAOeiBd

[John Com @JohnComFootball]

@forward_palk Yea v8. Without team detection I think I estimated 5-6 hours. With team detection 3 minutes took 30 minutes. I have to think there are ways to optimize team detection though. Also ML/CV isn’t exactly my area of expertise

[John Com @JohnComFootball]

I can't get over that fact that maybe 5-10 years ago, something like this was probably impossible. Now, if I wanted, I could probably whipped this together in an hour 🤯🤯🤯

Imagine having this kind of tracking data at every level of the game...

[John Com @JohnComFootball]

⚽️ Soccer Vision October! Day 5
Excited to share my first tracking clip! Found a random MLS goal, threw it at my code and 30s later I had this video!

It's far from perfect. Players flicker back and forth between "Player" and "Referee", but I think think that is in-part from this video being a different angle and distance from the training data. Had the trained the model with MLS footage, I'd imagine I'd have better results.