Realtime heat 2 · Cool★ 5heat incomplete (no X)

jev-box-tracking

Realtime object tracking that fuses slow Jev Box answers with per-frame optical flow in a Kalman filter, rewinding history when a late answer arrives.

What it is

A clean pattern for using a slow, accurate detector inside a fast loop.

What it does: Tracks a named object live; flow handles motion, Jev corrects absolute position.

Why it belongs in the showcase: Shows how to hide decision-model latency with classic filtering.

Caveats: Preview Box feature; per-call cost.

Evidence

Public Apache-2.0 Python package (5★ at intake) created 2026-10-16. README explains Lucas-Kanade + RANSAC flow inside the box between answers, a history buffer that rewinds the filter to frame k when its Jev box arrives, a `jev-box-track "iPhone"` webcam demo, and a `JevTracker` library whose `update()` never blocks on the network. Reads TYPESAFE_API_KEY from env.

Caveats

  • Box is a preview feature in Jev 1.14; accuracy varies with object size
  • One API round trip per Jev answer; cost scales with frame sampling
  • Small new repo; test on your own camera and targets

Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.