F FRAME / STREAM LAB
ON-DEVICE COMPUTE v2.0

CONTINUOUS VIDEO · BINARY COMPUTATION

Find the work that fits.

Eight retained sheets. One new frame at a time. Measure the complete update against a 16 ms budget.

DEVELOPMENT BENCHMARK
Untrained, deterministic BNN
01 Incoming pixels→02 Standardize→03 8-sheet stream→04 BNN + result
ReadyChoose an input and start a run
Complete update · p95— ms16 ms target
Observed source cadence— fpsCallback cadence — fps
Deadline misses—Updates over 16 ms
Skipped / gap events—Application + observed source

Update latency

Preprocessing → retained state → completed compute → result ready

Actual 16 ms
Your measurements will appear here.No estimated performance is displayed.

Standardized input

8 × Y
Awaiting input sheet

256 × 144 × 8 = 294,912 space–time samples

The preview is the current brightness sheet. Eight sheets are temporal input, not eight network layers.

Measurement boundary

16 ms
History span at 60 fps
116.67 ms
New sheet interval
16.67 ms
In-flight updates
1 · no growing queue
Startup / shader compile
Reported separately
First-result latency
Reported separately
Warmup results
0 / 120 completed
Sensor-frame loss
Not fully observable

A synthetic pass measures this browser's processing capacity. It does not qualify camera delivery or another device.

Capacity and comparisons

MAXIMUM / 16 ms

callback-through-facade-result/2 · UI on · preview on

Compare complete-update maxima and logical binary work. Earlier v1 controller results retain a different measurement boundary.

No campaign started. Full campaigns freeze candidates after exploration; failed confirmations remain visible.

Physical Android and iPhone qualification: pending device access and explicit capture. Hosting: static package; deployment status is separate.

Run ledger

Run / workloadBackendp95MaximumMissesOutcome
Runs and failures remain visible throughout this session.

Raw timing receipts are archived in this browser between runs. Stream state remains volatile; no video pixels are retained. Export before clearing browser storage.

Ready for local measurement. BNN weights are synthetic; this suite makes no detection-accuracy or clinical claim.