Created a Live Poker Player Database Using Computer Vision
Machine learning background here, new-ish to the poker scene. Was listening to Negreanu on Lex Fridman's podcast & was surprised to hear that live poker statistics/databases are relatively underdeveloped (aside from basic VPIP/PFR).
I've been messing around with a computer vision model to capture a frame-by-frame output of YouTube livestreams (pics attached) + while a bit buggy, it's doing the trick.
I created a spreadsheet capturing all 213 hands of sample HCL livestream that I'm linking here. Once this frame-by-frame data is captured, creating statistics / tendencies is relatively simple. Hopefully this sort of thing can be useful for pro players looking to get an edge.
https://docs.google.com/spreadsheets/d/1...
Feel confident I can get this to production ready & running on many different livestreams. Feel free to reach at [email]pokercomputervision@gmail.com[/email] if interested. Don't believe something like this has been created before but could be wrong.
5 Replies
I think Matt Berkey at solve4why has been trying to develop something similar for years.
It would be great if you could generate hand histories to download so we can import/ analyze them. Are you considering that?
It would be great if you could generate hand histories to download so we can import/ analyze them. Are you considering that?
Yep. Exported the sample HCL game to Excel as linked in original post but working to export in HH file format.
As you can imagine though, computing 1000s of hours of game footage can run up costs, so looking to eventually sell access. Happy to make it free for a bit.
Pretty cool. I was actually thinking of doing something like this for myself. I'm a dev but never worked with computer vision. I figured if we could track sports(soccer, bball, etc.) we should be able to track a poker livestream too.
