← sources

Looking for a Computer Vision technical co-founder (equity)

source: reddit (reddit:computervision)published 2026-11-09 14:02ingested 2026-11-10 02:37

retained text

I've spent almost two years and a significant amount of my own money building a football analytics product. I'm a software/data engineer, not a computer vision specialist, and I think I've reached the limits of what I can realistically accomplish alone.

Here's what I've already built in C++:

* **Camera/pitch calibration:** Dynamic homography estimation that follows a moving broadcast camera, keeping the pitch markings aligned as the camera pans and zooms.
* **Player detection and tracking at 25 Hz:** Relatively stable IDs while players remain visible, even when two or three players cross paths or move close together.
* **Consistent on-screen tracking:** The trajectories match what happens within the camera's field of view. I haven't prioritized trajectory smoothing because more fundamental problems remain unsolved.

**The two major challenges left:**

1. Robust player Re-ID, maintaining identities across occlusions, camera movement, and players leaving/re-entering the frame.
2. Associating jersey numbers with player identities reliably.

I also have access to **professional broadcast footage, purchased tracking datasets, and structured ball-event data**. I've invested heavily in acquiring these resources.  
  
At this point, I've hit a plateau. I've invested nearly two years of my time and money, and I honestly don't have the financial resources or energy to keep tackling such a difficult research problem alone.

I'm looking for someone experienced in computer vision, particularly MOT, Re-ID, temporal modeling, or sports analytics, who's interested in joining as a **technical partner in exchange for equity**. This isn't a salaried position at the moment.

I genuinely believe this has serious commercial potential.

And one thing I want to emphasize: **this isn't another weekend vibe-coded sports tracking demo claiming that football computer vision is solved.**

Building something that works consistently across real broadcast matches is incredibly difficult. At least it has been for me. There's a massive difference between producing an impressive demo and developing a reliable product that professional clubs can actually use.

If you've worked on similar problems and find this challenge interesting, I'd love to connect.

**DM me and let's talk.**

document fetched 2026-11-10 00:23 · canonical https://reddit.com/r/computervision/comments/1xrsfr4/looking_for_a_computer_vision_technical_cofounder