Pre-pilot · recruiting design-partner gyms
Grappling footage,
finally searchable.
Martial Vision is a coach- and athlete-governed research initiative building trustworthy computer vision for grappling. Every roll becomes a verified timeline. Every timeline becomes evidence.
The Platform
Film review that answers questions.
Not an AI coach. A workspace that turns raw rolls into structured, searchable evidence — with a coach checking the system's work at every step.
- 01
Upload
Phone or fixed-camera footage. No special hardware, ever.
- 02
Draft timeline
Broad positions and transitions, proposed automatically.
- 03
Coach review
Uncertain moments get flagged. Corrections take seconds — and every one makes the system better.
- 04
Search & clip
"Show every time John lost half guard." Click. Jump. Export for tomorrow's class.
- 05
Patterns
Athlete and team trends across months — where practice time actually goes.
Spend less time scrubbing footage. More time coaching from evidence.
The Problem
Grappling film is evidence coaches can't use.
Every practice produces footage. Almost none of it gets watched. Finding one lost underhook from last month means scrubbing hours of video by hand. Other sports built entire film-analytics industries. Grappling got skipped.
It got skipped because it's brutally hard. Limbs disappear under bodies. Generic pose models break down exactly where grappling begins: occlusion, contact, and context.
And the labels that matter — half guard, crossface, underhook — are martial-arts concepts, not computer-vision concepts. No public model knows them well.
What's missing isn't another AI demo. It's the foundation: a shared vocabulary, coach-verified data with clean rights, and honest benchmarks measured on footage the models have never seen.
A position isn't a shape.
It's a relationship between two bodies.
Scope, Honestly
What we are not building.
Narrow scope is a feature, not an apology.
- ×Hardware
No custom cameras, sensors, or wearables.
- ×Striking analysis
Boxing and stand-up analytics are someone else's fight.
- ×Scoring & judging
We don't score fights.
- ×Live cornering
Post-session analysis only.
- ×"Best move" advice
Video shows what happened — not what would have happened. We won't pretend otherwise.
- ×Medical claims
No fatigue, injury, damage, or heart-rate estimation. Ever.
Research Principles
How we keep ourselves honest.
Coach in the loop
The people who understand fighting best aren't software engineers. Coaches verify everything the system claims — and their corrections are what make it smarter.
Honest uncertainty
We'd rather show a coach "42% confident — needs your eyes" than pretend to be right all the time. The system knows what it doesn't know.
Real evaluation
Models are tested on unseen athletes, sessions, cameras, and gyms. Accuracy that only works on familiar footage is fake accuracy.
Open where it counts
The ontology, annotation format, and findings are open. Identifiable footage and private gym data are controlled. Sparring film is never public by default.
Privacy & Participant Rights
Your footage. Your call.
Training film exposes strategies, weaknesses, and bodies. Every athlete consents individually — gym-owner permission alone is not enough. Every participant picks a level:
Level 1
Private
Used only for your gym's own review. Nothing leaves. Never used for training or research.
Level 2
Anonymous research
Footage and labels help build models and aggregate findings. Nothing released publicly.
Level 3
Public contribution
Consent-cleared clips may join an open dataset or benchmark — credited or anonymous, your choice.
- Withdraw at any time; raw footage deleted on request.
- Research use only during the pilot — commercial use would require new, explicit opt-in.
- Encrypted storage; access limited to you and your gym's coaches.
- We will never infer pain, injury, fatigue, or any medical or psychological state.
Roadmap
Small claims, kept — in order.
-
Foundation Now
Mission, scope, consent framework, coach and athlete interviews. Recruit the first design-partner gyms.
-
Ontology & manual pilot
Define the position and transition vocabulary. Run the workflow with humans in the loop and measure whether it actually saves coaches time.
-
AI assistance
Automate suggestions only where they demonstrably cut review time. Coach corrections keep improving the system.
-
Cross-gym benchmark
Blind evaluation on unseen athletes, environments, and cameras. Publish what works and what doesn't.
-
Longitudinal analytics & broader grappling
Athlete patterns, team trends, curriculum support. Then — and only then — gi BJJ, wrestling, judo, and MMA ground work.
Get Involved
This only works as a collaboration.
Coaches
Shape the tool from day one. Tell us what film review costs you today, help define the vocabulary, and correct the system's mistakes — your expertise is the product.
Athletes
See your own patterns across months of training — with full control over your footage and how it's used.
Gyms
Become a design partner. Participating gyms get early access to every tool the initiative develops, and a real say in what gets built.
Researchers & developers
Grappling is a genuinely hard, underserved vision problem. Help build the ontology, benchmarks, and baselines the field is missing.
Contact
Skeptical? Curious?
Both are welcome.
hello@martialvision.example
Replace with your real address before publishing.