AI Inference Benchmark

Real-Time JS Demo · 3-Stage Latency · QoE Metrics · TensorFlow.js

Model

Backend

REC

—

Starting…

Pred. ms

—

FPS

—

JS Heap MB

—

Warmup ms

—

Setup ms

—

QoE Metrics

Responsiveness

—

runs/min
60000 ÷ lat

Smoothness

—

fps
live render

Infer. Acc.

—

avg conf.
[0 – 1]

Metrics reported independently — no composite score

—

Select a model to begin.

This Frame

Prediction ms

FPS

Detections

Responsiveness (runs/min)

Smoothness (fps)

JS Heap MB

3-Stage Latency Breakdown

Per model × backend: Setup (load) · Warmup (first inference) · Prediction (steady-state avg/min/max/σ).
Switch backends to populate multiple rows.

Setup — model load & framework init

Warmup — first inference, shader compile

Prediction — steady-state inference

Run a model to populate this table.

Comparison Summary

By Model (all backends combined)

Run each model to populate.

By Backend

Switch backends to populate.

Session Log 0 samples

#	Time	Model	Task	Backend	Pred ms	FPS	Det	Resp	Smth	IAcc	Heap MB
Press ⏺ Record then switch models and backends.