Results

Results from the local pipeline.

A quick look at what the Mac produces today: overlays, run metrics, and the local review workflow around recorded sports clips.

Positioning

RefereAI is the local-first lane.

Product Best for Runs on Pricing / access Workflow
RefereAI MLX Local observer + HIL review loop Mac on local network Open-source, self-hosted Analyze clips, generate overlays, review mistakes, keep private footage local
SwingVision Tennis match analytics Phone + cloud sync Consumer subscription Capture and upload a tennis match library
Hawk-Eye Elite officiating Venue install Enterprise contract Calibrated venue operations for top-tier events
PlaySight / Hudl Team video platforms Cloud Team / facility plans Organize, share, and analyze team video libraries
PB Vision Pickleball review Cloud Sport-specific app access Upload-first pickleball analysis workflow
Sports coverage

Built across court, net, and gym sports.

The classifier is intentionally multi-sport. Volleyball is the strongest public demo lane today; tennis, table tennis, and badminton have solid checked-in references; pickleball and basketball are active improvement lanes.

Volleyball

Hero lane for public demos, indoor gym clips, team separation, and rally overlays.

Tennis

Racket and court evidence with strong local reference clips.

Table tennis

Small-court/table geometry plus paddle-size cues.

Badminton

Long-racket and court-shape signals for indoor play.

Pickleball

Tracked as an active lane for paddle/court disambiguation.

Basketball

Gym-ball and player-pattern lane for open-gym scenes.

Run metrics

yolo26n on checked-in clips.

37.7 FPS

tennis_rally.mp4 · 0.82 confidence · 37.8 ms p95.

34.5 FPS

volleyball_larochette · volleyball · 0.70 confidence.

58.8 FPS

badminton_clip.mp4 · badminton · 0.60 confidence.

Zero false positives

Known misses are left as unknown instead of pretending certainty.

The checked-in harness runs the same pipeline: video → YOLO26 MLX → tracker → sport classifier → overlay video. Larger local engines can be compared with refereai compare when installed.

Workflow

From clip to review queue.

RefereAI MLX focuses on sport inference, overlays, session artifacts, and a human review loop for messy amateur sports footage.

The sweet spot is fast local understanding: take a clip or phone stream, produce reviewable evidence, and let a person correct the system when the overlay is wrong.