Spaces:
Running
Running
Upload 2 files
Browse files- README.md +3 -3
- index.html +452 -373
README.md
CHANGED
|
@@ -5,11 +5,11 @@ colorFrom: blue
|
|
| 5 |
colorTo: purple
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
| 8 |
-
short_description: Live
|
| 9 |
---
|
| 10 |
|
| 11 |
# SentAI
|
| 12 |
|
| 13 |
-
|
| 14 |
|
| 15 |
-
Phase
|
|
|
|
| 5 |
colorTo: purple
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
| 8 |
+
short_description: Live face emotion age gender estimates
|
| 9 |
---
|
| 10 |
|
| 11 |
# SentAI
|
| 12 |
|
| 13 |
+
Simple browser-side live face analysis for Hugging Face Spaces.
|
| 14 |
|
| 15 |
+
Phase 4 removes manual model and accuracy controls. Precision models load automatically in the background. The app estimates visible facial expression, apparent age range, and male/female presentation from live camera frames.
|
index.html
CHANGED
|
@@ -6,6 +6,7 @@
|
|
| 6 |
<meta name="theme-color" content="#050816" />
|
| 7 |
<title>SentAI</title>
|
| 8 |
<script defer src="https://cdn.jsdelivr.net/npm/face-api.js@0.22.2/dist/face-api.min.js"></script>
|
|
|
|
| 9 |
<style>
|
| 10 |
:root {
|
| 11 |
--bg-a: #050816;
|
|
@@ -69,18 +70,18 @@
|
|
| 69 |
|
| 70 |
.brand h1 {
|
| 71 |
margin: 0;
|
| 72 |
-
font-size: clamp(
|
| 73 |
-
line-height: 0.
|
| 74 |
letter-spacing: -0.08em;
|
| 75 |
font-weight: 950;
|
| 76 |
-
text-shadow: 0 24px 70px rgba(34, 211, 238, 0.
|
| 77 |
}
|
| 78 |
|
| 79 |
.brand p {
|
| 80 |
margin: 18px 0 0;
|
| 81 |
-
max-width:
|
| 82 |
color: var(--muted);
|
| 83 |
-
font-size: clamp(1rem, 1.
|
| 84 |
line-height: 1.55;
|
| 85 |
}
|
| 86 |
|
|
@@ -102,7 +103,7 @@
|
|
| 102 |
color: var(--muted);
|
| 103 |
box-shadow: var(--shadow);
|
| 104 |
white-space: nowrap;
|
| 105 |
-
font-weight:
|
| 106 |
}
|
| 107 |
|
| 108 |
.dot {
|
|
@@ -124,7 +125,7 @@
|
|
| 124 |
margin-bottom: 18px;
|
| 125 |
}
|
| 126 |
|
| 127 |
-
button
|
| 128 |
appearance: none;
|
| 129 |
border: 1px solid var(--stroke);
|
| 130 |
background: rgba(15, 23, 42, 0.76);
|
|
@@ -139,10 +140,9 @@
|
|
| 139 |
min-height: 48px;
|
| 140 |
}
|
| 141 |
|
| 142 |
-
button:hover
|
| 143 |
button:active { transform: translateY(0); }
|
| 144 |
button.primary { background: linear-gradient(135deg, rgba(34,211,238,0.96), rgba(167,139,250,0.94)); color: #06111f; border-color: transparent; }
|
| 145 |
-
button.accent { color: #cffafe; border-color: rgba(34, 211, 238, 0.4); }
|
| 146 |
button.danger { color: #fecdd3; }
|
| 147 |
button:disabled { opacity: 0.46; cursor: not-allowed; transform: none; }
|
| 148 |
|
|
@@ -160,7 +160,7 @@
|
|
| 160 |
|
| 161 |
.grid {
|
| 162 |
display: grid;
|
| 163 |
-
grid-template-columns: minmax(0, 1.
|
| 164 |
gap: 18px;
|
| 165 |
align-items: stretch;
|
| 166 |
}
|
|
@@ -211,9 +211,9 @@
|
|
| 211 |
}
|
| 212 |
|
| 213 |
.empty-card {
|
| 214 |
-
max-width:
|
| 215 |
border: 1px solid var(--stroke);
|
| 216 |
-
background: rgba(15,23,42,0.
|
| 217 |
border-radius: 24px;
|
| 218 |
padding: 28px;
|
| 219 |
}
|
|
@@ -278,14 +278,14 @@
|
|
| 278 |
|
| 279 |
.wide { grid-column: 1 / -1; }
|
| 280 |
|
| 281 |
-
.bars, .note
|
| 282 |
border: 1px solid var(--stroke);
|
| 283 |
background: rgba(255,255,255,0.055);
|
| 284 |
border-radius: 22px;
|
| 285 |
padding: 16px;
|
| 286 |
}
|
| 287 |
|
| 288 |
-
.bars h3
|
| 289 |
margin: 0 0 14px;
|
| 290 |
font-size: 1rem;
|
| 291 |
color: var(--text);
|
|
@@ -318,21 +318,6 @@
|
|
| 318 |
transition: width 180ms ease;
|
| 319 |
}
|
| 320 |
|
| 321 |
-
.calibration-row {
|
| 322 |
-
display: grid;
|
| 323 |
-
grid-template-columns: 90px 1fr 60px;
|
| 324 |
-
gap: 12px;
|
| 325 |
-
align-items: center;
|
| 326 |
-
color: var(--muted);
|
| 327 |
-
font-weight: 800;
|
| 328 |
-
margin-top: 10px;
|
| 329 |
-
}
|
| 330 |
-
|
| 331 |
-
input[type="range"] {
|
| 332 |
-
width: 100%;
|
| 333 |
-
accent-color: var(--accent);
|
| 334 |
-
}
|
| 335 |
-
|
| 336 |
.note {
|
| 337 |
color: #cffafe;
|
| 338 |
background: rgba(34,211,238,0.08);
|
|
@@ -371,15 +356,14 @@
|
|
| 371 |
.stage-panel { padding: 10px; }
|
| 372 |
.side { padding: 14px; }
|
| 373 |
.toolbar { gap: 9px; }
|
| 374 |
-
button,
|
| 375 |
}
|
| 376 |
|
| 377 |
@media (max-width: 560px) {
|
| 378 |
-
.brand h1 { font-size: clamp(
|
| 379 |
.brand p { font-size: 0.98rem; }
|
| 380 |
.metric-grid { grid-template-columns: 1fr; }
|
| 381 |
.bar-row { grid-template-columns: 76px 1fr 44px; font-size: 0.82rem; }
|
| 382 |
-
.calibration-row { grid-template-columns: 78px 1fr 52px; }
|
| 383 |
.empty-card { padding: 20px; }
|
| 384 |
.app-shell { width: min(100% - 14px, 760px); }
|
| 385 |
body { background-attachment: fixed; }
|
|
@@ -391,29 +375,18 @@
|
|
| 391 |
<header class="hero" aria-label="SentAI heading">
|
| 392 |
<div class="brand">
|
| 393 |
<h1>SentAI</h1>
|
| 394 |
-
<p>
|
| 395 |
</div>
|
| 396 |
<div class="status-stack">
|
| 397 |
-
<div class="status-pill" aria-live="polite"><span id="
|
| 398 |
-
<div class="status-pill" aria-live="polite"><span id="
|
| 399 |
</div>
|
| 400 |
</header>
|
| 401 |
|
| 402 |
<section class="toolbar" aria-label="Camera controls">
|
| 403 |
<button id="startBtn" class="primary" disabled>Start camera</button>
|
| 404 |
<button id="switchBtn" disabled>Switch front/rear</button>
|
| 405 |
-
<button id="accuracyBtn" class="accent" disabled>Load accuracy pack</button>
|
| 406 |
<button id="stopBtn" class="danger" disabled>Stop</button>
|
| 407 |
-
<select id="emotionMode" aria-label="Emotion scoring mode">
|
| 408 |
-
<option value="sensitive" selected>Boost sad/fear/disgust</option>
|
| 409 |
-
<option value="balanced">Balanced emotions</option>
|
| 410 |
-
<option value="raw">Raw model scores</option>
|
| 411 |
-
</select>
|
| 412 |
-
<select id="modeSelect" aria-label="Performance mode">
|
| 413 |
-
<option value="fast">Fast mode</option>
|
| 414 |
-
<option value="balanced">Balanced mode</option>
|
| 415 |
-
<option value="accurate" selected>Accurate mode</option>
|
| 416 |
-
</select>
|
| 417 |
<span id="cameraTag" class="camera-tag">Camera: not started</span>
|
| 418 |
</section>
|
| 419 |
|
|
@@ -425,7 +398,7 @@
|
|
| 425 |
<div id="emptyState" class="empty-state">
|
| 426 |
<div class="empty-card">
|
| 427 |
<strong>Ready for live analysis</strong>
|
| 428 |
-
Tap <b>Start camera</b>.
|
| 429 |
</div>
|
| 430 |
</div>
|
| 431 |
</div>
|
|
@@ -436,17 +409,17 @@
|
|
| 436 |
<div class="metric-grid">
|
| 437 |
<div class="metric">
|
| 438 |
<span>Possible feeling</span>
|
| 439 |
-
<strong id="feelingValue">
|
| 440 |
<small id="feelingConfidence">Waiting</small>
|
| 441 |
</div>
|
| 442 |
<div class="metric">
|
| 443 |
<span>Gender</span>
|
| 444 |
-
<strong id="genderValue">
|
| 445 |
<small id="genderConfidence">Waiting</small>
|
| 446 |
</div>
|
| 447 |
<div class="metric">
|
| 448 |
<span>Apparent age</span>
|
| 449 |
-
<strong id="ageValue">
|
| 450 |
<small id="ageSource">Waiting</small>
|
| 451 |
</div>
|
| 452 |
<div class="metric">
|
|
@@ -456,13 +429,13 @@
|
|
| 456 |
</div>
|
| 457 |
<div class="metric">
|
| 458 |
<span>FPS</span>
|
| 459 |
-
<strong id="fpsValue">
|
| 460 |
-
<small>
|
| 461 |
</div>
|
| 462 |
<div class="metric">
|
| 463 |
<span>Latency</span>
|
| 464 |
-
<strong id="latencyValue">
|
| 465 |
-
<small id="latencySource">
|
| 466 |
</div>
|
| 467 |
</div>
|
| 468 |
|
|
@@ -471,17 +444,8 @@
|
|
| 471 |
<div id="emotionBars"></div>
|
| 472 |
</div>
|
| 473 |
|
| 474 |
-
<div class="calibration wide">
|
| 475 |
-
<h3>Age fine tune</h3>
|
| 476 |
-
<div class="calibration-row">
|
| 477 |
-
<span>Offset</span>
|
| 478 |
-
<input id="ageOffset" type="range" min="-12" max="12" value="0" step="1" aria-label="Age offset calibration" />
|
| 479 |
-
<b id="ageOffsetValue">0y</b>
|
| 480 |
-
</div>
|
| 481 |
-
</div>
|
| 482 |
-
|
| 483 |
<div class="note wide">
|
| 484 |
-
|
| 485 |
</div>
|
| 486 |
</aside>
|
| 487 |
</section>
|
|
@@ -491,22 +455,21 @@
|
|
| 491 |
|
| 492 |
<script type="module">
|
| 493 |
import { pipeline as xenovaPipeline, env as xenovaEnv } from "https://cdn.jsdelivr.net/npm/@xenova/transformers@2.17.2";
|
|
|
|
|
|
|
| 494 |
const FACE_API_MODEL_URL = "https://cdn.jsdelivr.net/gh/justadudewhohacks/face-api.js@0.22.2/weights";
|
| 495 |
-
const
|
| 496 |
-
const
|
| 497 |
-
const
|
| 498 |
|
| 499 |
const els = {
|
| 500 |
-
|
| 501 |
-
|
| 502 |
-
|
| 503 |
-
|
| 504 |
startBtn: document.getElementById("startBtn"),
|
| 505 |
switchBtn: document.getElementById("switchBtn"),
|
| 506 |
-
accuracyBtn: document.getElementById("accuracyBtn"),
|
| 507 |
stopBtn: document.getElementById("stopBtn"),
|
| 508 |
-
emotionMode: document.getElementById("emotionMode"),
|
| 509 |
-
modeSelect: document.getElementById("modeSelect"),
|
| 510 |
cameraTag: document.getElementById("cameraTag"),
|
| 511 |
video: document.getElementById("video"),
|
| 512 |
overlay: document.getElementById("overlay"),
|
|
@@ -524,18 +487,12 @@
|
|
| 524 |
latencyValue: document.getElementById("latencyValue"),
|
| 525 |
latencySource: document.getElementById("latencySource"),
|
| 526 |
emotionBars: document.getElementById("emotionBars"),
|
| 527 |
-
ageOffset: document.getElementById("ageOffset"),
|
| 528 |
-
ageOffsetValue: document.getElementById("ageOffsetValue"),
|
| 529 |
toast: document.getElementById("toast"),
|
| 530 |
};
|
| 531 |
|
| 532 |
-
const modes = {
|
| 533 |
-
fast: { inputSize: 224, scoreThreshold: 0.52, interval: 22, proInterval: 2800, smoothing: 0.18 },
|
| 534 |
-
balanced: { inputSize: 320, scoreThreshold: 0.45, interval: 42, proInterval: 2100, smoothing: 0.24 },
|
| 535 |
-
accurate: { inputSize: 416, scoreThreshold: 0.38, interval: 65, proInterval: 1550, smoothing: 0.30 },
|
| 536 |
-
};
|
| 537 |
-
|
| 538 |
const emotionLabels = ["Happy", "Sad", "Fear", "Anger", "Confused", "Disgust"];
|
|
|
|
|
|
|
| 539 |
const ctx = els.overlay.getContext("2d");
|
| 540 |
|
| 541 |
let coreReady = false;
|
|
@@ -549,26 +506,28 @@
|
|
| 549 |
let emotionEma = null;
|
| 550 |
let genderMaleEma = null;
|
| 551 |
let lastBaseAgeSample = 0;
|
|
|
|
| 552 |
|
| 553 |
const ageHistory = [];
|
| 554 |
-
const
|
| 555 |
loading: false,
|
| 556 |
-
|
|
|
|
| 557 |
emotionPipe: null,
|
| 558 |
-
|
| 559 |
-
|
| 560 |
-
loadImage: null,
|
| 561 |
lastRun: 0,
|
| 562 |
busy: false,
|
| 563 |
-
|
| 564 |
-
|
|
|
|
|
|
|
| 565 |
age: null,
|
| 566 |
ageAt: 0,
|
| 567 |
gender: null,
|
| 568 |
genderScore: 0,
|
| 569 |
genderAt: 0,
|
| 570 |
latency: 0,
|
| 571 |
-
device: "wasm",
|
| 572 |
};
|
| 573 |
|
| 574 |
function setPill(dot, label, text, kind = "loading") {
|
|
@@ -582,7 +541,7 @@
|
|
| 582 |
els.toast.textContent = message;
|
| 583 |
els.toast.classList.add("show");
|
| 584 |
clearTimeout(showToast.timer);
|
| 585 |
-
showToast.timer = setTimeout(() => els.toast.classList.remove("show"),
|
| 586 |
}
|
| 587 |
|
| 588 |
function clamp01(value) {
|
|
@@ -597,6 +556,15 @@
|
|
| 597 |
return 1 / (1 + Math.exp(-x));
|
| 598 |
}
|
| 599 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 600 |
function median(values) {
|
| 601 |
const list = values.filter(Number.isFinite).slice().sort((a, b) => a - b);
|
| 602 |
if (!list.length) return NaN;
|
|
@@ -604,17 +572,6 @@
|
|
| 604 |
return list.length % 2 ? list[mid] : (list[mid - 1] + list[mid]) / 2;
|
| 605 |
}
|
| 606 |
|
| 607 |
-
function ageRange(age, source = "core", samples = 1) {
|
| 608 |
-
if (!Number.isFinite(age)) return "—";
|
| 609 |
-
const offset = Number(els.ageOffset.value || 0);
|
| 610 |
-
const corrected = Math.max(0, Math.min(100, age + offset));
|
| 611 |
-
const half = source === "accuracy pack" ? (samples >= 3 ? 4 : 5) : (samples >= 6 ? 6 : 8);
|
| 612 |
-
const lo = Math.max(0, Math.round(corrected - half));
|
| 613 |
-
const hi = Math.min(100, Math.round(corrected + half));
|
| 614 |
-
if (hi <= 12) return "0-12";
|
| 615 |
-
return `${lo}-${hi}`;
|
| 616 |
-
}
|
| 617 |
-
|
| 618 |
function blankScores() {
|
| 619 |
return Object.fromEntries(emotionLabels.map(label => [label, 0]));
|
| 620 |
}
|
|
@@ -625,6 +582,24 @@
|
|
| 625 |
return out;
|
| 626 |
}
|
| 627 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 628 |
function calibrateFaceApiExpressions(expressions = {}) {
|
| 629 |
const raw = {
|
| 630 |
happy: clamp01(expressions.happy),
|
|
@@ -635,24 +610,21 @@
|
|
| 635 |
surprised: clamp01(expressions.surprised),
|
| 636 |
neutral: clamp01(expressions.neutral),
|
| 637 |
};
|
| 638 |
-
|
| 639 |
const nonNeutralTop = Math.max(raw.happy, raw.sad, raw.fearful, raw.angry, raw.disgusted, raw.surprised);
|
| 640 |
-
const
|
| 641 |
-
|
| 642 |
return normalizeScores({
|
| 643 |
-
Happy: Math.pow(raw.happy, 1.
|
| 644 |
-
Sad: Math.pow(raw.sad,
|
| 645 |
-
Fear: Math.max(Math.pow(raw.fearful,
|
| 646 |
-
Anger: Math.pow(raw.angry, 1.
|
| 647 |
-
Disgust: Math.pow(raw.disgusted,
|
| 648 |
-
Confused:
|
| 649 |
});
|
| 650 |
}
|
| 651 |
|
| 652 |
function normalizeExternalEmotion(outputs) {
|
| 653 |
const scores = blankScores();
|
| 654 |
const list = Array.isArray(outputs) ? outputs : [outputs];
|
| 655 |
-
|
| 656 |
for (const item of list) {
|
| 657 |
const label = String(item.label || item.class || "").toLowerCase();
|
| 658 |
const score = clamp01(item.score || item.probability || 0);
|
|
@@ -662,85 +634,80 @@
|
|
| 662 |
else if (label.includes("angry") || label.includes("anger")) scores.Anger = Math.max(scores.Anger, score);
|
| 663 |
else if (label.includes("disgust") || label.includes("disgusted")) scores.Disgust = Math.max(scores.Disgust, score);
|
| 664 |
else if (label.includes("surprise") || label.includes("neutral")) {
|
| 665 |
-
|
| 666 |
-
|
| 667 |
-
scores.Confused = Math.max(scores.Confused, Math.min(0.30, scaled));
|
| 668 |
}
|
| 669 |
}
|
| 670 |
return normalizeScores(scores);
|
| 671 |
}
|
| 672 |
|
| 673 |
-
function
|
| 674 |
-
const
|
| 675 |
-
if (mode === "raw") {
|
| 676 |
-
return { Happy: 1.00, Sad: 1.00, Fear: 1.00, Anger: 1.00, Confused: 1.00, Disgust: 1.00 };
|
| 677 |
-
}
|
| 678 |
-
if (mode === "balanced") {
|
| 679 |
-
return { Happy: 0.88, Sad: 1.28, Fear: 1.38, Anger: 0.95, Confused: 0.58, Disgust: 1.48 };
|
| 680 |
-
}
|
| 681 |
-
// Default: compensate for webcam models over-predicting smile/neutral/anger and under-predicting subtle negative expressions.
|
| 682 |
-
return { Happy: 0.70, Sad: 1.62, Fear: 1.82, Anger: 0.86, Confused: 0.42, Disgust: 2.05 };
|
| 683 |
-
}
|
| 684 |
-
|
| 685 |
-
function applyEmotionCalibration(scores) {
|
| 686 |
-
const mode = els.emotionMode?.value || "sensitive";
|
| 687 |
-
if (mode === "raw") return normalizeScores(scores);
|
| 688 |
-
const weights = emotionWeights();
|
| 689 |
const out = blankScores();
|
| 690 |
for (const label of emotionLabels) {
|
| 691 |
let v = clamp01(scores[label] || 0);
|
| 692 |
-
|
| 693 |
-
if (
|
| 694 |
-
if (label === "
|
| 695 |
-
|
| 696 |
-
out[label] = clamp01(v * weights[label]);
|
| 697 |
}
|
| 698 |
-
|
| 699 |
-
out.Confused = Math.min(out.Confused, mode === "balanced" ? 0.34 : 0.24);
|
| 700 |
-
return normalizeScores(out);
|
| 701 |
}
|
| 702 |
|
| 703 |
function combineEmotionScores(faceScores) {
|
| 704 |
const now = performance.now();
|
| 705 |
-
const
|
|
|
|
| 706 |
const combined = blankScores();
|
| 707 |
-
|
| 708 |
-
|
| 709 |
-
|
| 710 |
-
|
| 711 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 712 |
}
|
| 713 |
|
| 714 |
-
|
| 715 |
-
|
|
|
|
|
|
|
| 716 |
if (!emotionEma) {
|
| 717 |
emotionEma = calibrated;
|
| 718 |
} else {
|
| 719 |
for (const label of emotionLabels) {
|
| 720 |
-
emotionEma[label] = emotionEma[label] * (1 -
|
| 721 |
}
|
| 722 |
}
|
| 723 |
-
return
|
| 724 |
}
|
| 725 |
|
| 726 |
function topEmotion(scores) {
|
| 727 |
-
const
|
| 728 |
-
let [label, score] =
|
| 729 |
-
const
|
| 730 |
-
const
|
| 731 |
-
|
| 732 |
-
|
| 733 |
-
|
| 734 |
-
|
| 735 |
-
score = rare[1];
|
| 736 |
-
}
|
| 737 |
}
|
| 738 |
-
if (score < 0.
|
| 739 |
-
label = "Confused";
|
| 740 |
-
score = Math.max(scores.Confused || 0, 0.16);
|
| 741 |
-
} else if ((scores.Confused || 0) > score && score < 0.26) {
|
| 742 |
label = "Confused";
|
| 743 |
-
score = Math.
|
| 744 |
}
|
| 745 |
return { label, score: clamp01(score) };
|
| 746 |
}
|
|
@@ -749,14 +716,14 @@
|
|
| 749 |
if (!Number.isFinite(age) || age < 0 || age > 100) return;
|
| 750 |
const now = performance.now();
|
| 751 |
ageHistory.push({ age, source, weight, t: now });
|
| 752 |
-
while (ageHistory.length >
|
| 753 |
-
const cutoff = now -
|
| 754 |
while (ageHistory.length && ageHistory[0].t < cutoff) ageHistory.shift();
|
| 755 |
}
|
| 756 |
|
| 757 |
function stableAgeEstimate() {
|
| 758 |
-
const
|
| 759 |
-
const usable =
|
| 760 |
if (!usable.length) return { age: NaN, source: "Waiting", samples: 0 };
|
| 761 |
const expanded = [];
|
| 762 |
for (const sample of usable) {
|
|
@@ -765,7 +732,7 @@
|
|
| 765 |
}
|
| 766 |
return {
|
| 767 |
age: median(expanded),
|
| 768 |
-
source:
|
| 769 |
samples: usable.length,
|
| 770 |
};
|
| 771 |
}
|
|
@@ -773,12 +740,12 @@
|
|
| 773 |
function updateGenderEstimate(label, confidence, weight = 1) {
|
| 774 |
if (!label) return;
|
| 775 |
const pMale = label.toLowerCase() === "male" ? clamp01(confidence) : 1 - clamp01(confidence);
|
| 776 |
-
const alpha = Math.min(0.
|
| 777 |
genderMaleEma = genderMaleEma === null ? pMale : genderMaleEma * (1 - alpha) + pMale * alpha;
|
| 778 |
}
|
| 779 |
|
| 780 |
function currentGender() {
|
| 781 |
-
if (genderMaleEma === null) return { label: "
|
| 782 |
const label = genderMaleEma >= 0.5 ? "Male" : "Female";
|
| 783 |
const confidence = Math.max(genderMaleEma, 1 - genderMaleEma);
|
| 784 |
return { label, confidence };
|
|
@@ -790,20 +757,17 @@
|
|
| 790 |
const emotion = topEmotion(scores);
|
| 791 |
|
| 792 |
const now = performance.now();
|
| 793 |
-
if (Number.isFinite(det.age) && now - lastBaseAgeSample >
|
| 794 |
pushAgeSample(det.age, "core model", 1);
|
| 795 |
lastBaseAgeSample = now;
|
| 796 |
}
|
| 797 |
|
| 798 |
const coreGender = (det.gender || "").toLowerCase() === "female" ? "Female" : "Male";
|
| 799 |
updateGenderEstimate(coreGender, det.genderProbability || 0, 1);
|
| 800 |
-
if (
|
| 801 |
-
updateGenderEstimate(pro.gender, pro.genderScore, 3.2);
|
| 802 |
-
}
|
| 803 |
|
| 804 |
const age = stableAgeEstimate();
|
| 805 |
const gender = currentGender();
|
| 806 |
-
|
| 807 |
return {
|
| 808 |
emotionLabel: emotion.label,
|
| 809 |
emotionScore: emotion.score,
|
|
@@ -832,16 +796,16 @@
|
|
| 832 |
emotionEma = null;
|
| 833 |
genderMaleEma = null;
|
| 834 |
ageHistory.length = 0;
|
| 835 |
-
els.feelingValue.textContent = "
|
| 836 |
els.feelingConfidence.textContent = "Waiting";
|
| 837 |
-
els.genderValue.textContent = "
|
| 838 |
els.genderConfidence.textContent = "Waiting";
|
| 839 |
-
els.ageValue.textContent = "
|
| 840 |
els.ageSource.textContent = "Waiting";
|
| 841 |
els.facesValue.textContent = "0";
|
| 842 |
els.faceScore.textContent = "No face yet";
|
| 843 |
-
els.latencyValue.textContent = "
|
| 844 |
-
els.latencySource.textContent = "
|
| 845 |
renderBars({});
|
| 846 |
}
|
| 847 |
|
|
@@ -854,10 +818,21 @@
|
|
| 854 |
})[0];
|
| 855 |
}
|
| 856 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 857 |
function updateDetails(detections, elapsedMs) {
|
| 858 |
els.facesValue.textContent = String(detections.length);
|
| 859 |
els.latencyValue.textContent = `${Math.round(elapsedMs)}ms`;
|
| 860 |
-
els.latencySource.textContent =
|
| 861 |
|
| 862 |
const now = performance.now();
|
| 863 |
const instantFps = 1000 / Math.max(1, now - lastLoopTime);
|
|
@@ -866,24 +841,25 @@
|
|
| 866 |
els.fpsValue.textContent = fpsSmooth.toFixed(1);
|
| 867 |
|
| 868 |
const primary = choosePrimary(detections);
|
|
|
|
| 869 |
if (!primary) {
|
| 870 |
-
els.feelingValue.textContent = "
|
| 871 |
els.feelingConfidence.textContent = "No face";
|
| 872 |
-
els.genderValue.textContent = "
|
| 873 |
els.genderConfidence.textContent = "No face";
|
| 874 |
-
els.ageValue.textContent = "
|
| 875 |
els.ageSource.textContent = "No face";
|
| 876 |
els.faceScore.textContent = "No face yet";
|
| 877 |
renderBars({});
|
| 878 |
return;
|
| 879 |
}
|
| 880 |
|
| 881 |
-
|
| 882 |
const insight = makeInsight(primary);
|
| 883 |
els.feelingValue.textContent = insight.emotionLabel;
|
| 884 |
els.feelingConfidence.textContent = `${percent(insight.emotionScore)} confidence`;
|
| 885 |
els.genderValue.textContent = insight.gender;
|
| 886 |
-
els.genderConfidence.textContent = insight.gender === "
|
| 887 |
els.ageValue.textContent = insight.ageRange;
|
| 888 |
els.ageSource.textContent = insight.ageSource;
|
| 889 |
els.faceScore.textContent = `${percent(insight.faceScore)} face score`;
|
|
@@ -930,7 +906,7 @@
|
|
| 930 |
const font2 = Math.round(14 * scale);
|
| 931 |
const lineH = font1 + 9 * scale;
|
| 932 |
const label1 = `${insight.emotionLabel} ${percent(insight.emotionScore)}`;
|
| 933 |
-
const label2 = `${insight.gender} ${percent(insight.genderScore)}
|
| 934 |
|
| 935 |
ctx.font = `900 ${font1}px Inter, system-ui, sans-serif`;
|
| 936 |
const textW = Math.max(ctx.measureText(label1).width, ctx.measureText(label2).width);
|
|
@@ -980,7 +956,7 @@
|
|
| 980 |
if (lastDetections.length) drawDetections(lastDetections);
|
| 981 |
}
|
| 982 |
|
| 983 |
-
function cropFaceCanvas(det, targetSize =
|
| 984 |
const box = det.detection.box;
|
| 985 |
const videoW = els.video.videoWidth || els.overlay.width;
|
| 986 |
const videoH = els.video.videoHeight || els.overlay.height;
|
|
@@ -1020,228 +996,323 @@
|
|
| 1020 |
});
|
| 1021 |
}
|
| 1022 |
|
| 1023 |
-
async function
|
| 1024 |
const started = performance.now();
|
| 1025 |
-
while (!window
|
| 1026 |
-
if (performance.now() - started >
|
| 1027 |
await new Promise(resolve => setTimeout(resolve, 80));
|
| 1028 |
}
|
|
|
|
| 1029 |
}
|
| 1030 |
|
| 1031 |
async function loadCoreModels() {
|
| 1032 |
try {
|
| 1033 |
-
await
|
| 1034 |
await Promise.all([
|
|
|
|
| 1035 |
faceapi.nets.tinyFaceDetector.loadFromUri(FACE_API_MODEL_URL),
|
| 1036 |
faceapi.nets.faceLandmark68TinyNet.loadFromUri(FACE_API_MODEL_URL),
|
| 1037 |
faceapi.nets.faceExpressionNet.loadFromUri(FACE_API_MODEL_URL),
|
| 1038 |
faceapi.nets.ageGenderNet.loadFromUri(FACE_API_MODEL_URL),
|
| 1039 |
]);
|
| 1040 |
coreReady = true;
|
| 1041 |
-
setPill(els.
|
| 1042 |
els.startBtn.disabled = false;
|
| 1043 |
-
els.accuracyBtn.disabled = false;
|
| 1044 |
renderBars({});
|
| 1045 |
-
|
| 1046 |
} catch (err) {
|
| 1047 |
console.error(err);
|
| 1048 |
-
setPill(els.
|
| 1049 |
-
showToast("
|
| 1050 |
}
|
| 1051 |
}
|
| 1052 |
|
| 1053 |
-
async function
|
| 1054 |
-
|
| 1055 |
-
|
| 1056 |
-
|
| 1057 |
-
|
| 1058 |
-
|
| 1059 |
-
mod.env.useBrowserCache = true;
|
| 1060 |
-
mod.env.backends ??= {};
|
| 1061 |
-
mod.env.backends.onnx ??= {};
|
| 1062 |
-
mod.env.backends.onnx.wasm ??= {};
|
| 1063 |
-
mod.env.backends.onnx.wasm.numThreads = Math.max(1, Math.min(4, navigator.hardwareConcurrency || 2));
|
| 1064 |
-
}
|
| 1065 |
-
return mod;
|
| 1066 |
}
|
| 1067 |
|
| 1068 |
-
async function
|
| 1069 |
-
|
| 1070 |
-
|
| 1071 |
-
|
| 1072 |
-
{}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1073 |
];
|
| 1074 |
-
|
| 1075 |
-
|
| 1076 |
-
|
| 1077 |
-
|
| 1078 |
-
|
| 1079 |
-
|
| 1080 |
-
|
| 1081 |
-
|
| 1082 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1083 |
}
|
| 1084 |
}
|
| 1085 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1086 |
}
|
| 1087 |
|
| 1088 |
-
|
| 1089 |
-
const
|
| 1090 |
-
const
|
| 1091 |
-
|
| 1092 |
-
|
| 1093 |
-
let
|
| 1094 |
-
|
| 1095 |
-
|
| 1096 |
-
|
| 1097 |
-
|
| 1098 |
-
pro.device = opts.device || "wasm";
|
| 1099 |
-
break;
|
| 1100 |
-
} catch (err) {
|
| 1101 |
-
lastErr = err;
|
| 1102 |
-
console.warn("Age/gender model load attempt failed", opts, err);
|
| 1103 |
-
}
|
| 1104 |
}
|
| 1105 |
-
|
| 1106 |
-
const processor = await mod.AutoProcessor.from_pretrained(AGE_GENDER_MODEL_ID);
|
| 1107 |
-
return { model, processor };
|
| 1108 |
}
|
| 1109 |
|
| 1110 |
-
async function
|
| 1111 |
-
if (
|
| 1112 |
-
|
| 1113 |
-
|
| 1114 |
-
|
| 1115 |
-
|
| 1116 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1117 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1118 |
try {
|
| 1119 |
-
const
|
| 1120 |
-
|
| 1121 |
-
|
| 1122 |
-
|
| 1123 |
-
|
| 1124 |
-
|
| 1125 |
-
|
| 1126 |
-
|
| 1127 |
-
|
| 1128 |
-
try {
|
| 1129 |
-
if (AGE_GENDER_MODEL_ID && mod.AutoModel && mod.AutoProcessor && (mod.load_image || mod.RawImage?.fromURL)) {
|
| 1130 |
-
const pair = await loadAgeGenderWithFallback(mod);
|
| 1131 |
-
pro.ageGenderModel = pair.model;
|
| 1132 |
-
pro.ageGenderProcessor = pair.processor;
|
| 1133 |
}
|
| 1134 |
-
|
| 1135 |
-
|
|
|
|
| 1136 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1137 |
|
| 1138 |
-
|
| 1139 |
-
|
| 1140 |
-
|
| 1141 |
-
|
| 1142 |
-
|
| 1143 |
-
|
| 1144 |
-
|
| 1145 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1146 |
}
|
| 1147 |
-
} catch (err) {
|
| 1148 |
-
console.error(err);
|
| 1149 |
-
setPill(els.proDot, els.proStatus, "Emotion accuracy pack unavailable; using core model", "error");
|
| 1150 |
-
if (!auto) showToast("Accuracy pack could not load. The app will keep using the core model.");
|
| 1151 |
} finally {
|
| 1152 |
-
|
| 1153 |
-
els.accuracyBtn.disabled = !!(pro.emotionPipe && pro.ageGenderModel);
|
| 1154 |
}
|
| 1155 |
}
|
| 1156 |
|
| 1157 |
-
async function
|
| 1158 |
-
const cfg = modes[els.modeSelect.value] || modes.accurate;
|
| 1159 |
const now = performance.now();
|
| 1160 |
-
if (!primary ||
|
| 1161 |
-
if (now -
|
| 1162 |
-
|
| 1163 |
-
|
| 1164 |
-
|
| 1165 |
-
let url = null;
|
| 1166 |
const started = performance.now();
|
| 1167 |
try {
|
| 1168 |
-
|
| 1169 |
-
|
| 1170 |
-
|
| 1171 |
-
|
| 1172 |
-
|
| 1173 |
-
cropFaceCanvas(primary, 288, 0.16, "contrast(1.10) saturate(0.96)", false),
|
| 1174 |
-
cropFaceCanvas(primary, 288, 0.34, "contrast(1.18) saturate(0.92)", false),
|
| 1175 |
-
cropFaceCanvas(primary, 288, 0.06, "contrast(1.22) brightness(1.03)", false),
|
| 1176 |
-
cropFaceCanvas(primary, 288, 0.22, "contrast(1.14) saturate(0.92)", true),
|
| 1177 |
-
];
|
| 1178 |
-
const urls = [];
|
| 1179 |
-
const aggregate = blankScores();
|
| 1180 |
-
let count = 0;
|
| 1181 |
-
try {
|
| 1182 |
-
for (const variant of cropVariants) {
|
| 1183 |
-
const variantUrl = await canvasToBlobUrl(variant);
|
| 1184 |
-
urls.push(variantUrl);
|
| 1185 |
-
let output;
|
| 1186 |
-
try {
|
| 1187 |
-
output = await pro.emotionPipe(variantUrl, { topK: 7 });
|
| 1188 |
-
} catch (_) {
|
| 1189 |
-
output = await pro.emotionPipe(variantUrl);
|
| 1190 |
-
}
|
| 1191 |
-
const scores = normalizeExternalEmotion(output);
|
| 1192 |
-
for (const label of emotionLabels) aggregate[label] += scores[label] || 0;
|
| 1193 |
-
count += 1;
|
| 1194 |
-
}
|
| 1195 |
-
} finally {
|
| 1196 |
-
for (const variantUrl of urls) URL.revokeObjectURL(variantUrl);
|
| 1197 |
}
|
| 1198 |
-
for (const label of emotionLabels) aggregate[label] = count ? aggregate[label] / count : 0;
|
| 1199 |
-
pro.emotionScores = normalizeScores(aggregate);
|
| 1200 |
-
pro.emotionAt = performance.now();
|
| 1201 |
}
|
| 1202 |
-
|
| 1203 |
-
|
| 1204 |
-
|
| 1205 |
-
|
| 1206 |
-
|
| 1207 |
-
}
|
| 1208 |
-
const inputs = await pro.ageGenderProcessor(image);
|
| 1209 |
-
const output = await pro.ageGenderModel(inputs);
|
| 1210 |
-
const logits = output.logits || output.last_hidden_state || output[0];
|
| 1211 |
-
let values = [];
|
| 1212 |
-
if (logits?.tolist) {
|
| 1213 |
-
values = logits.tolist().flat(Infinity);
|
| 1214 |
-
} else if (logits?.data) {
|
| 1215 |
-
values = Array.from(logits.data);
|
| 1216 |
-
}
|
| 1217 |
-
const rawAge = Number(values[0]);
|
| 1218 |
-
const rawGender = Number(values[1]);
|
| 1219 |
-
if (Number.isFinite(rawAge)) {
|
| 1220 |
-
const age = Math.max(0, Math.min(100, Math.round(rawAge)));
|
| 1221 |
-
pro.age = age;
|
| 1222 |
-
pro.ageAt = performance.now();
|
| 1223 |
-
pushAgeSample(age, "accuracy pack", 4);
|
| 1224 |
-
}
|
| 1225 |
-
if (Number.isFinite(rawGender)) {
|
| 1226 |
-
const pFemale = rawGender >= 0 && rawGender <= 1 ? rawGender : sigmoid(rawGender);
|
| 1227 |
-
pro.gender = pFemale >= 0.5 ? "Female" : "Male";
|
| 1228 |
-
pro.genderScore = Math.max(pFemale, 1 - pFemale);
|
| 1229 |
-
pro.genderAt = performance.now();
|
| 1230 |
}
|
| 1231 |
}
|
|
|
|
| 1232 |
} catch (err) {
|
| 1233 |
-
console.warn("
|
| 1234 |
} finally {
|
| 1235 |
-
|
| 1236 |
-
|
| 1237 |
-
pro.busy = false;
|
| 1238 |
}
|
| 1239 |
}
|
| 1240 |
|
| 1241 |
function stopStream() {
|
| 1242 |
-
if (stream)
|
| 1243 |
-
for (const track of stream.getTracks()) track.stop();
|
| 1244 |
-
}
|
| 1245 |
stream = null;
|
| 1246 |
running = false;
|
| 1247 |
detecting = false;
|
|
@@ -1257,7 +1328,7 @@
|
|
| 1257 |
|
| 1258 |
async function startCamera(facing = currentFacing) {
|
| 1259 |
if (!coreReady) {
|
| 1260 |
-
showToast("Please wait until the
|
| 1261 |
return;
|
| 1262 |
}
|
| 1263 |
|
|
@@ -1269,7 +1340,7 @@
|
|
| 1269 |
audio: false,
|
| 1270 |
video: {
|
| 1271 |
facingMode: { ideal: currentFacing },
|
| 1272 |
-
width: { ideal:
|
| 1273 |
height: { ideal: 720 },
|
| 1274 |
}
|
| 1275 |
};
|
|
@@ -1302,19 +1373,32 @@
|
|
| 1302 |
detectLoop();
|
| 1303 |
}
|
| 1304 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1305 |
async function detectLoop() {
|
| 1306 |
if (!running || !coreReady || detecting) return;
|
| 1307 |
detecting = true;
|
| 1308 |
const started = performance.now();
|
| 1309 |
try {
|
| 1310 |
fitCanvas();
|
| 1311 |
-
const
|
| 1312 |
-
const options = new faceapi.TinyFaceDetectorOptions({ inputSize: cfg.inputSize, scoreThreshold: cfg.scoreThreshold });
|
| 1313 |
-
const raw = await faceapi
|
| 1314 |
-
.detectAllFaces(els.video, options)
|
| 1315 |
-
.withFaceLandmarks(true)
|
| 1316 |
-
.withFaceExpressions()
|
| 1317 |
-
.withAgeAndGender();
|
| 1318 |
const displaySize = { width: els.overlay.width, height: els.overlay.height };
|
| 1319 |
const resized = faceapi.resizeResults(raw, displaySize);
|
| 1320 |
const elapsed = performance.now() - started;
|
|
@@ -1323,7 +1407,7 @@
|
|
| 1323 |
setTimeout(() => {
|
| 1324 |
detecting = false;
|
| 1325 |
requestAnimationFrame(detectLoop);
|
| 1326 |
-
},
|
| 1327 |
} catch (err) {
|
| 1328 |
console.error(err);
|
| 1329 |
detecting = false;
|
|
@@ -1337,18 +1421,13 @@
|
|
| 1337 |
currentFacing = currentFacing === "user" ? "environment" : "user";
|
| 1338 |
startCamera(currentFacing);
|
| 1339 |
});
|
| 1340 |
-
els.accuracyBtn.addEventListener("click", () => loadAccuracyPack(false));
|
| 1341 |
els.stopBtn.addEventListener("click", stopStream);
|
| 1342 |
els.video.addEventListener("loadedmetadata", fitCanvas);
|
| 1343 |
window.addEventListener("resize", fitCanvas);
|
| 1344 |
-
els.ageOffset.addEventListener("input", () => {
|
| 1345 |
-
const value = Number(els.ageOffset.value || 0);
|
| 1346 |
-
els.ageOffsetValue.textContent = `${value > 0 ? "+" : ""}${value}y`;
|
| 1347 |
-
});
|
| 1348 |
|
| 1349 |
resetDetails();
|
| 1350 |
renderBars({});
|
| 1351 |
-
setPill(els.
|
| 1352 |
loadCoreModels();
|
| 1353 |
</script>
|
| 1354 |
</body>
|
|
|
|
| 6 |
<meta name="theme-color" content="#050816" />
|
| 7 |
<title>SentAI</title>
|
| 8 |
<script defer src="https://cdn.jsdelivr.net/npm/face-api.js@0.22.2/dist/face-api.min.js"></script>
|
| 9 |
+
<script defer src="https://cdn.jsdelivr.net/npm/onnxruntime-web@1.21.0/dist/ort.min.js"></script>
|
| 10 |
<style>
|
| 11 |
:root {
|
| 12 |
--bg-a: #050816;
|
|
|
|
| 70 |
|
| 71 |
.brand h1 {
|
| 72 |
margin: 0;
|
| 73 |
+
font-size: clamp(4.2rem, 9vw, 8rem);
|
| 74 |
+
line-height: 0.88;
|
| 75 |
letter-spacing: -0.08em;
|
| 76 |
font-weight: 950;
|
| 77 |
+
text-shadow: 0 24px 70px rgba(34, 211, 238, 0.10);
|
| 78 |
}
|
| 79 |
|
| 80 |
.brand p {
|
| 81 |
margin: 18px 0 0;
|
| 82 |
+
max-width: 960px;
|
| 83 |
color: var(--muted);
|
| 84 |
+
font-size: clamp(1rem, 1.7vw, 1.22rem);
|
| 85 |
line-height: 1.55;
|
| 86 |
}
|
| 87 |
|
|
|
|
| 103 |
color: var(--muted);
|
| 104 |
box-shadow: var(--shadow);
|
| 105 |
white-space: nowrap;
|
| 106 |
+
font-weight: 850;
|
| 107 |
}
|
| 108 |
|
| 109 |
.dot {
|
|
|
|
| 125 |
margin-bottom: 18px;
|
| 126 |
}
|
| 127 |
|
| 128 |
+
button {
|
| 129 |
appearance: none;
|
| 130 |
border: 1px solid var(--stroke);
|
| 131 |
background: rgba(15, 23, 42, 0.76);
|
|
|
|
| 140 |
min-height: 48px;
|
| 141 |
}
|
| 142 |
|
| 143 |
+
button:hover { transform: translateY(-1px); border-color: rgba(34, 211, 238, 0.55); }
|
| 144 |
button:active { transform: translateY(0); }
|
| 145 |
button.primary { background: linear-gradient(135deg, rgba(34,211,238,0.96), rgba(167,139,250,0.94)); color: #06111f; border-color: transparent; }
|
|
|
|
| 146 |
button.danger { color: #fecdd3; }
|
| 147 |
button:disabled { opacity: 0.46; cursor: not-allowed; transform: none; }
|
| 148 |
|
|
|
|
| 160 |
|
| 161 |
.grid {
|
| 162 |
display: grid;
|
| 163 |
+
grid-template-columns: minmax(0, 1.42fr) minmax(360px, 0.76fr);
|
| 164 |
gap: 18px;
|
| 165 |
align-items: stretch;
|
| 166 |
}
|
|
|
|
| 211 |
}
|
| 212 |
|
| 213 |
.empty-card {
|
| 214 |
+
max-width: 620px;
|
| 215 |
border: 1px solid var(--stroke);
|
| 216 |
+
background: rgba(15,23,42,0.70);
|
| 217 |
border-radius: 24px;
|
| 218 |
padding: 28px;
|
| 219 |
}
|
|
|
|
| 278 |
|
| 279 |
.wide { grid-column: 1 / -1; }
|
| 280 |
|
| 281 |
+
.bars, .note {
|
| 282 |
border: 1px solid var(--stroke);
|
| 283 |
background: rgba(255,255,255,0.055);
|
| 284 |
border-radius: 22px;
|
| 285 |
padding: 16px;
|
| 286 |
}
|
| 287 |
|
| 288 |
+
.bars h3 {
|
| 289 |
margin: 0 0 14px;
|
| 290 |
font-size: 1rem;
|
| 291 |
color: var(--text);
|
|
|
|
| 318 |
transition: width 180ms ease;
|
| 319 |
}
|
| 320 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 321 |
.note {
|
| 322 |
color: #cffafe;
|
| 323 |
background: rgba(34,211,238,0.08);
|
|
|
|
| 356 |
.stage-panel { padding: 10px; }
|
| 357 |
.side { padding: 14px; }
|
| 358 |
.toolbar { gap: 9px; }
|
| 359 |
+
button, .camera-tag { flex: 1 1 150px; justify-content: center; }
|
| 360 |
}
|
| 361 |
|
| 362 |
@media (max-width: 560px) {
|
| 363 |
+
.brand h1 { font-size: clamp(4rem, 22vw, 5.4rem); }
|
| 364 |
.brand p { font-size: 0.98rem; }
|
| 365 |
.metric-grid { grid-template-columns: 1fr; }
|
| 366 |
.bar-row { grid-template-columns: 76px 1fr 44px; font-size: 0.82rem; }
|
|
|
|
| 367 |
.empty-card { padding: 20px; }
|
| 368 |
.app-shell { width: min(100% - 14px, 760px); }
|
| 369 |
body { background-attachment: fixed; }
|
|
|
|
| 375 |
<header class="hero" aria-label="SentAI heading">
|
| 376 |
<div class="brand">
|
| 377 |
<h1>SentAI</h1>
|
| 378 |
+
<p>Live facial analysis with automatic high-precision models for expression, apparent age range, and male/female estimate. No manual model switches.</p>
|
| 379 |
</div>
|
| 380 |
<div class="status-stack">
|
| 381 |
+
<div class="status-pill" aria-live="polite"><span id="aiDot" class="dot"></span><span id="aiStatus">Preparing AI models...</span></div>
|
| 382 |
+
<div class="status-pill" aria-live="polite"><span id="precisionDot" class="dot"></span><span id="precisionStatus">Precision models loading automatically</span></div>
|
| 383 |
</div>
|
| 384 |
</header>
|
| 385 |
|
| 386 |
<section class="toolbar" aria-label="Camera controls">
|
| 387 |
<button id="startBtn" class="primary" disabled>Start camera</button>
|
| 388 |
<button id="switchBtn" disabled>Switch front/rear</button>
|
|
|
|
| 389 |
<button id="stopBtn" class="danger" disabled>Stop</button>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 390 |
<span id="cameraTag" class="camera-tag">Camera: not started</span>
|
| 391 |
</section>
|
| 392 |
|
|
|
|
| 398 |
<div id="emptyState" class="empty-state">
|
| 399 |
<div class="empty-card">
|
| 400 |
<strong>Ready for live analysis</strong>
|
| 401 |
+
Tap <b>Start camera</b>. Use bright front lighting, keep one face centered, and hold each expression for a moment so the neural ensemble can stabilize.
|
| 402 |
</div>
|
| 403 |
</div>
|
| 404 |
</div>
|
|
|
|
| 409 |
<div class="metric-grid">
|
| 410 |
<div class="metric">
|
| 411 |
<span>Possible feeling</span>
|
| 412 |
+
<strong id="feelingValue">-</strong>
|
| 413 |
<small id="feelingConfidence">Waiting</small>
|
| 414 |
</div>
|
| 415 |
<div class="metric">
|
| 416 |
<span>Gender</span>
|
| 417 |
+
<strong id="genderValue">-</strong>
|
| 418 |
<small id="genderConfidence">Waiting</small>
|
| 419 |
</div>
|
| 420 |
<div class="metric">
|
| 421 |
<span>Apparent age</span>
|
| 422 |
+
<strong id="ageValue">-</strong>
|
| 423 |
<small id="ageSource">Waiting</small>
|
| 424 |
</div>
|
| 425 |
<div class="metric">
|
|
|
|
| 429 |
</div>
|
| 430 |
<div class="metric">
|
| 431 |
<span>FPS</span>
|
| 432 |
+
<strong id="fpsValue">-</strong>
|
| 433 |
+
<small>Live loop</small>
|
| 434 |
</div>
|
| 435 |
<div class="metric">
|
| 436 |
<span>Latency</span>
|
| 437 |
+
<strong id="latencyValue">-</strong>
|
| 438 |
+
<small id="latencySource">Model time</small>
|
| 439 |
</div>
|
| 440 |
</div>
|
| 441 |
|
|
|
|
| 444 |
<div id="emotionBars"></div>
|
| 445 |
</div>
|
| 446 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 447 |
<div class="note wide">
|
| 448 |
+
SentAI estimates visible facial expression and apparent age from camera frames. It cannot know a person's true internal feeling, but this version uses an automatic multi-model deep-learning ensemble to improve sad, fear, and disgust detection.
|
| 449 |
</div>
|
| 450 |
</aside>
|
| 451 |
</section>
|
|
|
|
| 455 |
|
| 456 |
<script type="module">
|
| 457 |
import { pipeline as xenovaPipeline, env as xenovaEnv } from "https://cdn.jsdelivr.net/npm/@xenova/transformers@2.17.2";
|
| 458 |
+
import { AutoModel, AutoProcessor, load_image, env as hfEnv } from "https://cdn.jsdelivr.net/npm/@huggingface/transformers@3.8.1";
|
| 459 |
+
|
| 460 |
const FACE_API_MODEL_URL = "https://cdn.jsdelivr.net/gh/justadudewhohacks/face-api.js@0.22.2/weights";
|
| 461 |
+
const EMOTION_TRANSFORMER_MODEL_ID = "Xenova/facial_emotions_image_detection";
|
| 462 |
+
const AGE_GENDER_MODEL_ID = "onnx-community/age-gender-prediction-ONNX";
|
| 463 |
+
const OPENCV_FER_MODEL_URL = "https://huggingface.co/opencv/facial_expression_recognition/resolve/main/facial_expression_recognition_mobilefacenet_2022july.onnx";
|
| 464 |
|
| 465 |
const els = {
|
| 466 |
+
aiDot: document.getElementById("aiDot"),
|
| 467 |
+
aiStatus: document.getElementById("aiStatus"),
|
| 468 |
+
precisionDot: document.getElementById("precisionDot"),
|
| 469 |
+
precisionStatus: document.getElementById("precisionStatus"),
|
| 470 |
startBtn: document.getElementById("startBtn"),
|
| 471 |
switchBtn: document.getElementById("switchBtn"),
|
|
|
|
| 472 |
stopBtn: document.getElementById("stopBtn"),
|
|
|
|
|
|
|
| 473 |
cameraTag: document.getElementById("cameraTag"),
|
| 474 |
video: document.getElementById("video"),
|
| 475 |
overlay: document.getElementById("overlay"),
|
|
|
|
| 487 |
latencyValue: document.getElementById("latencyValue"),
|
| 488 |
latencySource: document.getElementById("latencySource"),
|
| 489 |
emotionBars: document.getElementById("emotionBars"),
|
|
|
|
|
|
|
| 490 |
toast: document.getElementById("toast"),
|
| 491 |
};
|
| 492 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 493 |
const emotionLabels = ["Happy", "Sad", "Fear", "Anger", "Confused", "Disgust"];
|
| 494 |
+
const ferLabels = ["Anger", "Disgust", "Fear", "Happy", "Confused", "Sad", "Confused"];
|
| 495 |
+
const detectorConfig = { inputSize: 416, scoreThreshold: 0.32, intervalMs: 65, precisionIntervalMs: 650, smoothing: 0.26 };
|
| 496 |
const ctx = els.overlay.getContext("2d");
|
| 497 |
|
| 498 |
let coreReady = false;
|
|
|
|
| 506 |
let emotionEma = null;
|
| 507 |
let genderMaleEma = null;
|
| 508 |
let lastBaseAgeSample = 0;
|
| 509 |
+
let latestPrimary = null;
|
| 510 |
|
| 511 |
const ageHistory = [];
|
| 512 |
+
const precision = {
|
| 513 |
loading: false,
|
| 514 |
+
readyCount: 0,
|
| 515 |
+
ferSession: null,
|
| 516 |
emotionPipe: null,
|
| 517 |
+
ageModel: null,
|
| 518 |
+
ageProcessor: null,
|
|
|
|
| 519 |
lastRun: 0,
|
| 520 |
busy: false,
|
| 521 |
+
ferScores: null,
|
| 522 |
+
ferAt: 0,
|
| 523 |
+
transformerScores: null,
|
| 524 |
+
transformerAt: 0,
|
| 525 |
age: null,
|
| 526 |
ageAt: 0,
|
| 527 |
gender: null,
|
| 528 |
genderScore: 0,
|
| 529 |
genderAt: 0,
|
| 530 |
latency: 0,
|
|
|
|
| 531 |
};
|
| 532 |
|
| 533 |
function setPill(dot, label, text, kind = "loading") {
|
|
|
|
| 541 |
els.toast.textContent = message;
|
| 542 |
els.toast.classList.add("show");
|
| 543 |
clearTimeout(showToast.timer);
|
| 544 |
+
showToast.timer = setTimeout(() => els.toast.classList.remove("show"), 3400);
|
| 545 |
}
|
| 546 |
|
| 547 |
function clamp01(value) {
|
|
|
|
| 556 |
return 1 / (1 + Math.exp(-x));
|
| 557 |
}
|
| 558 |
|
| 559 |
+
function softmax(values) {
|
| 560 |
+
const list = Array.from(values || []).map(v => Number(v));
|
| 561 |
+
if (!list.length) return [];
|
| 562 |
+
const max = Math.max(...list);
|
| 563 |
+
const exps = list.map(v => Math.exp(v - max));
|
| 564 |
+
const sum = exps.reduce((a, b) => a + b, 0) || 1;
|
| 565 |
+
return exps.map(v => v / sum);
|
| 566 |
+
}
|
| 567 |
+
|
| 568 |
function median(values) {
|
| 569 |
const list = values.filter(Number.isFinite).slice().sort((a, b) => a - b);
|
| 570 |
if (!list.length) return NaN;
|
|
|
|
| 572 |
return list.length % 2 ? list[mid] : (list[mid - 1] + list[mid]) / 2;
|
| 573 |
}
|
| 574 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 575 |
function blankScores() {
|
| 576 |
return Object.fromEntries(emotionLabels.map(label => [label, 0]));
|
| 577 |
}
|
|
|
|
| 582 |
return out;
|
| 583 |
}
|
| 584 |
|
| 585 |
+
function renormalize(scores) {
|
| 586 |
+
const out = normalizeScores(scores);
|
| 587 |
+
const sum = emotionLabels.reduce((a, label) => a + out[label], 0);
|
| 588 |
+
if (sum <= 0) return out;
|
| 589 |
+
for (const label of emotionLabels) out[label] = out[label] / sum;
|
| 590 |
+
return out;
|
| 591 |
+
}
|
| 592 |
+
|
| 593 |
+
function ageRange(age, source = "core", samples = 1) {
|
| 594 |
+
if (!Number.isFinite(age)) return "-";
|
| 595 |
+
const corrected = Math.max(0, Math.min(100, age));
|
| 596 |
+
const half = source === "precision model" ? (samples >= 3 ? 4 : 5) : (samples >= 6 ? 6 : 8);
|
| 597 |
+
const lo = Math.max(0, Math.round(corrected - half));
|
| 598 |
+
const hi = Math.min(100, Math.round(corrected + half));
|
| 599 |
+
if (hi <= 12) return "0-12";
|
| 600 |
+
return `${lo}-${hi}`;
|
| 601 |
+
}
|
| 602 |
+
|
| 603 |
function calibrateFaceApiExpressions(expressions = {}) {
|
| 604 |
const raw = {
|
| 605 |
happy: clamp01(expressions.happy),
|
|
|
|
| 610 |
surprised: clamp01(expressions.surprised),
|
| 611 |
neutral: clamp01(expressions.neutral),
|
| 612 |
};
|
|
|
|
| 613 |
const nonNeutralTop = Math.max(raw.happy, raw.sad, raw.fearful, raw.angry, raw.disgusted, raw.surprised);
|
| 614 |
+
const confused = Math.min(0.38, raw.neutral * 0.24 + raw.surprised * 0.42 + Math.max(0, 1 - nonNeutralTop) * 0.05);
|
|
|
|
| 615 |
return normalizeScores({
|
| 616 |
+
Happy: Math.pow(raw.happy, 1.10),
|
| 617 |
+
Sad: Math.pow(raw.sad, 0.96),
|
| 618 |
+
Fear: Math.max(Math.pow(raw.fearful, 0.98), raw.surprised * raw.fearful * 0.55),
|
| 619 |
+
Anger: Math.pow(raw.angry, 1.02),
|
| 620 |
+
Disgust: Math.pow(raw.disgusted, 0.96),
|
| 621 |
+
Confused: confused,
|
| 622 |
});
|
| 623 |
}
|
| 624 |
|
| 625 |
function normalizeExternalEmotion(outputs) {
|
| 626 |
const scores = blankScores();
|
| 627 |
const list = Array.isArray(outputs) ? outputs : [outputs];
|
|
|
|
| 628 |
for (const item of list) {
|
| 629 |
const label = String(item.label || item.class || "").toLowerCase();
|
| 630 |
const score = clamp01(item.score || item.probability || 0);
|
|
|
|
| 634 |
else if (label.includes("angry") || label.includes("anger")) scores.Anger = Math.max(scores.Anger, score);
|
| 635 |
else if (label.includes("disgust") || label.includes("disgusted")) scores.Disgust = Math.max(scores.Disgust, score);
|
| 636 |
else if (label.includes("surprise") || label.includes("neutral")) {
|
| 637 |
+
const scaled = label.includes("neutral") ? score * 0.45 : score * 0.52;
|
| 638 |
+
scores.Confused = Math.max(scores.Confused, Math.min(0.58, scaled));
|
|
|
|
| 639 |
}
|
| 640 |
}
|
| 641 |
return normalizeScores(scores);
|
| 642 |
}
|
| 643 |
|
| 644 |
+
function calibrateFinalEmotion(scores) {
|
| 645 |
+
const priors = { Happy: 0.86, Sad: 1.18, Fear: 1.24, Anger: 0.96, Confused: 0.72, Disgust: 1.34 };
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 646 |
const out = blankScores();
|
| 647 |
for (const label of emotionLabels) {
|
| 648 |
let v = clamp01(scores[label] || 0);
|
| 649 |
+
if (["Sad", "Fear", "Disgust"].includes(label)) v = Math.pow(v, 0.92);
|
| 650 |
+
if (label === "Happy") v = Math.pow(v, 1.08);
|
| 651 |
+
if (label === "Confused") v = Math.pow(v, 1.04);
|
| 652 |
+
out[label] = clamp01(v * priors[label]);
|
|
|
|
| 653 |
}
|
| 654 |
+
return renormalize(out);
|
|
|
|
|
|
|
| 655 |
}
|
| 656 |
|
| 657 |
function combineEmotionScores(faceScores) {
|
| 658 |
const now = performance.now();
|
| 659 |
+
const ferFresh = precision.ferScores && (now - precision.ferAt < 3400);
|
| 660 |
+
const transformerFresh = precision.transformerScores && (now - precision.transformerAt < 5000);
|
| 661 |
const combined = blankScores();
|
| 662 |
+
|
| 663 |
+
let totalWeight = 0;
|
| 664 |
+
const addWeighted = (scores, weight) => {
|
| 665 |
+
if (!scores || weight <= 0) return;
|
| 666 |
+
totalWeight += weight;
|
| 667 |
+
for (const label of emotionLabels) combined[label] += (scores[label] || 0) * weight;
|
| 668 |
+
};
|
| 669 |
+
|
| 670 |
+
if (ferFresh && transformerFresh) {
|
| 671 |
+
addWeighted(precision.ferScores, 0.55);
|
| 672 |
+
addWeighted(precision.transformerScores, 0.34);
|
| 673 |
+
addWeighted(faceScores, 0.11);
|
| 674 |
+
} else if (ferFresh) {
|
| 675 |
+
addWeighted(precision.ferScores, 0.76);
|
| 676 |
+
addWeighted(faceScores, 0.24);
|
| 677 |
+
} else if (transformerFresh) {
|
| 678 |
+
addWeighted(precision.transformerScores, 0.78);
|
| 679 |
+
addWeighted(faceScores, 0.22);
|
| 680 |
+
} else {
|
| 681 |
+
addWeighted(faceScores, 1.0);
|
| 682 |
}
|
| 683 |
|
| 684 |
+
if (totalWeight > 0) {
|
| 685 |
+
for (const label of emotionLabels) combined[label] /= totalWeight;
|
| 686 |
+
}
|
| 687 |
+
const calibrated = calibrateFinalEmotion(combined);
|
| 688 |
if (!emotionEma) {
|
| 689 |
emotionEma = calibrated;
|
| 690 |
} else {
|
| 691 |
for (const label of emotionLabels) {
|
| 692 |
+
emotionEma[label] = emotionEma[label] * (1 - detectorConfig.smoothing) + calibrated[label] * detectorConfig.smoothing;
|
| 693 |
}
|
| 694 |
}
|
| 695 |
+
return renormalize(emotionEma);
|
| 696 |
}
|
| 697 |
|
| 698 |
function topEmotion(scores) {
|
| 699 |
+
const entries = Object.entries(scores).sort((a, b) => b[1] - a[1]);
|
| 700 |
+
let [label, score] = entries[0] || ["Confused", 0];
|
| 701 |
+
const nonConfused = entries.filter(([name]) => name !== "Confused");
|
| 702 |
+
const [altLabel, altScore] = nonConfused[0] || [label, score];
|
| 703 |
+
|
| 704 |
+
if (label === "Confused" && altScore >= 0.22 && (score - altScore) < 0.10) {
|
| 705 |
+
label = altLabel;
|
| 706 |
+
score = altScore;
|
|
|
|
|
|
|
| 707 |
}
|
| 708 |
+
if (score < 0.18) {
|
|
|
|
|
|
|
|
|
|
| 709 |
label = "Confused";
|
| 710 |
+
score = Math.max(scores.Confused || 0.16, 0.16);
|
| 711 |
}
|
| 712 |
return { label, score: clamp01(score) };
|
| 713 |
}
|
|
|
|
| 716 |
if (!Number.isFinite(age) || age < 0 || age > 100) return;
|
| 717 |
const now = performance.now();
|
| 718 |
ageHistory.push({ age, source, weight, t: now });
|
| 719 |
+
while (ageHistory.length > 50) ageHistory.shift();
|
| 720 |
+
const cutoff = now - 45000;
|
| 721 |
while (ageHistory.length && ageHistory[0].t < cutoff) ageHistory.shift();
|
| 722 |
}
|
| 723 |
|
| 724 |
function stableAgeEstimate() {
|
| 725 |
+
const precisionSamples = ageHistory.filter(s => s.source === "precision model");
|
| 726 |
+
const usable = precisionSamples.length >= 2 ? precisionSamples : ageHistory;
|
| 727 |
if (!usable.length) return { age: NaN, source: "Waiting", samples: 0 };
|
| 728 |
const expanded = [];
|
| 729 |
for (const sample of usable) {
|
|
|
|
| 732 |
}
|
| 733 |
return {
|
| 734 |
age: median(expanded),
|
| 735 |
+
source: precisionSamples.length >= 2 ? "precision model" : "core model",
|
| 736 |
samples: usable.length,
|
| 737 |
};
|
| 738 |
}
|
|
|
|
| 740 |
function updateGenderEstimate(label, confidence, weight = 1) {
|
| 741 |
if (!label) return;
|
| 742 |
const pMale = label.toLowerCase() === "male" ? clamp01(confidence) : 1 - clamp01(confidence);
|
| 743 |
+
const alpha = Math.min(0.68, 0.15 * weight);
|
| 744 |
genderMaleEma = genderMaleEma === null ? pMale : genderMaleEma * (1 - alpha) + pMale * alpha;
|
| 745 |
}
|
| 746 |
|
| 747 |
function currentGender() {
|
| 748 |
+
if (genderMaleEma === null) return { label: "-", confidence: 0 };
|
| 749 |
const label = genderMaleEma >= 0.5 ? "Male" : "Female";
|
| 750 |
const confidence = Math.max(genderMaleEma, 1 - genderMaleEma);
|
| 751 |
return { label, confidence };
|
|
|
|
| 757 |
const emotion = topEmotion(scores);
|
| 758 |
|
| 759 |
const now = performance.now();
|
| 760 |
+
if (Number.isFinite(det.age) && now - lastBaseAgeSample > 700) {
|
| 761 |
pushAgeSample(det.age, "core model", 1);
|
| 762 |
lastBaseAgeSample = now;
|
| 763 |
}
|
| 764 |
|
| 765 |
const coreGender = (det.gender || "").toLowerCase() === "female" ? "Female" : "Male";
|
| 766 |
updateGenderEstimate(coreGender, det.genderProbability || 0, 1);
|
| 767 |
+
if (precision.gender && (now - precision.genderAt < 9000)) updateGenderEstimate(precision.gender, precision.genderScore, 3.8);
|
|
|
|
|
|
|
| 768 |
|
| 769 |
const age = stableAgeEstimate();
|
| 770 |
const gender = currentGender();
|
|
|
|
| 771 |
return {
|
| 772 |
emotionLabel: emotion.label,
|
| 773 |
emotionScore: emotion.score,
|
|
|
|
| 796 |
emotionEma = null;
|
| 797 |
genderMaleEma = null;
|
| 798 |
ageHistory.length = 0;
|
| 799 |
+
els.feelingValue.textContent = "-";
|
| 800 |
els.feelingConfidence.textContent = "Waiting";
|
| 801 |
+
els.genderValue.textContent = "-";
|
| 802 |
els.genderConfidence.textContent = "Waiting";
|
| 803 |
+
els.ageValue.textContent = "-";
|
| 804 |
els.ageSource.textContent = "Waiting";
|
| 805 |
els.facesValue.textContent = "0";
|
| 806 |
els.faceScore.textContent = "No face yet";
|
| 807 |
+
els.latencyValue.textContent = "-";
|
| 808 |
+
els.latencySource.textContent = "Model time";
|
| 809 |
renderBars({});
|
| 810 |
}
|
| 811 |
|
|
|
|
| 818 |
})[0];
|
| 819 |
}
|
| 820 |
|
| 821 |
+
function updatePrecisionStatus() {
|
| 822 |
+
const parts = [];
|
| 823 |
+
if (precision.ferSession) parts.push("MobileFaceNet emotion");
|
| 824 |
+
if (precision.emotionPipe) parts.push("ViT emotion");
|
| 825 |
+
if (precision.ageModel && precision.ageProcessor) parts.push("ViT age/gender");
|
| 826 |
+
if (parts.length >= 2) setPill(els.precisionDot, els.precisionStatus, `Precision ready: ${parts.join(" + ")}`, "ready");
|
| 827 |
+
else if (parts.length === 1) setPill(els.precisionDot, els.precisionStatus, `Precision partially ready: ${parts[0]}`, "ready");
|
| 828 |
+
else if (precision.loading) setPill(els.precisionDot, els.precisionStatus, "Precision models loading automatically", "loading");
|
| 829 |
+
else setPill(els.precisionDot, els.precisionStatus, "Precision models unavailable; using core AI", "error");
|
| 830 |
+
}
|
| 831 |
+
|
| 832 |
function updateDetails(detections, elapsedMs) {
|
| 833 |
els.facesValue.textContent = String(detections.length);
|
| 834 |
els.latencyValue.textContent = `${Math.round(elapsedMs)}ms`;
|
| 835 |
+
els.latencySource.textContent = precision.latency ? `Core + precision ${Math.round(precision.latency)}ms` : "Core model time";
|
| 836 |
|
| 837 |
const now = performance.now();
|
| 838 |
const instantFps = 1000 / Math.max(1, now - lastLoopTime);
|
|
|
|
| 841 |
els.fpsValue.textContent = fpsSmooth.toFixed(1);
|
| 842 |
|
| 843 |
const primary = choosePrimary(detections);
|
| 844 |
+
latestPrimary = primary;
|
| 845 |
if (!primary) {
|
| 846 |
+
els.feelingValue.textContent = "-";
|
| 847 |
els.feelingConfidence.textContent = "No face";
|
| 848 |
+
els.genderValue.textContent = "-";
|
| 849 |
els.genderConfidence.textContent = "No face";
|
| 850 |
+
els.ageValue.textContent = "-";
|
| 851 |
els.ageSource.textContent = "No face";
|
| 852 |
els.faceScore.textContent = "No face yet";
|
| 853 |
renderBars({});
|
| 854 |
return;
|
| 855 |
}
|
| 856 |
|
| 857 |
+
maybeRunPrecision(primary);
|
| 858 |
const insight = makeInsight(primary);
|
| 859 |
els.feelingValue.textContent = insight.emotionLabel;
|
| 860 |
els.feelingConfidence.textContent = `${percent(insight.emotionScore)} confidence`;
|
| 861 |
els.genderValue.textContent = insight.gender;
|
| 862 |
+
els.genderConfidence.textContent = insight.gender === "-" ? "Waiting" : `${percent(insight.genderScore)} confidence`;
|
| 863 |
els.ageValue.textContent = insight.ageRange;
|
| 864 |
els.ageSource.textContent = insight.ageSource;
|
| 865 |
els.faceScore.textContent = `${percent(insight.faceScore)} face score`;
|
|
|
|
| 906 |
const font2 = Math.round(14 * scale);
|
| 907 |
const lineH = font1 + 9 * scale;
|
| 908 |
const label1 = `${insight.emotionLabel} ${percent(insight.emotionScore)}`;
|
| 909 |
+
const label2 = `${insight.gender} ${percent(insight.genderScore)} | Age ${insight.ageRange}`;
|
| 910 |
|
| 911 |
ctx.font = `900 ${font1}px Inter, system-ui, sans-serif`;
|
| 912 |
const textW = Math.max(ctx.measureText(label1).width, ctx.measureText(label2).width);
|
|
|
|
| 956 |
if (lastDetections.length) drawDetections(lastDetections);
|
| 957 |
}
|
| 958 |
|
| 959 |
+
function cropFaceCanvas(det, targetSize = 288, pad = 0.28, filter = "none", mirror = false) {
|
| 960 |
const box = det.detection.box;
|
| 961 |
const videoW = els.video.videoWidth || els.overlay.width;
|
| 962 |
const videoH = els.video.videoHeight || els.overlay.height;
|
|
|
|
| 996 |
});
|
| 997 |
}
|
| 998 |
|
| 999 |
+
async function waitForGlobal(name, timeoutMs = 14000) {
|
| 1000 |
const started = performance.now();
|
| 1001 |
+
while (!window[name]) {
|
| 1002 |
+
if (performance.now() - started > timeoutMs) throw new Error(`${name} did not load`);
|
| 1003 |
await new Promise(resolve => setTimeout(resolve, 80));
|
| 1004 |
}
|
| 1005 |
+
return window[name];
|
| 1006 |
}
|
| 1007 |
|
| 1008 |
async function loadCoreModels() {
|
| 1009 |
try {
|
| 1010 |
+
await waitForGlobal("faceapi");
|
| 1011 |
await Promise.all([
|
| 1012 |
+
faceapi.nets.ssdMobilenetv1.loadFromUri(FACE_API_MODEL_URL),
|
| 1013 |
faceapi.nets.tinyFaceDetector.loadFromUri(FACE_API_MODEL_URL),
|
| 1014 |
faceapi.nets.faceLandmark68TinyNet.loadFromUri(FACE_API_MODEL_URL),
|
| 1015 |
faceapi.nets.faceExpressionNet.loadFromUri(FACE_API_MODEL_URL),
|
| 1016 |
faceapi.nets.ageGenderNet.loadFromUri(FACE_API_MODEL_URL),
|
| 1017 |
]);
|
| 1018 |
coreReady = true;
|
| 1019 |
+
setPill(els.aiDot, els.aiStatus, "AI ready", "ready");
|
| 1020 |
els.startBtn.disabled = false;
|
|
|
|
| 1021 |
renderBars({});
|
| 1022 |
+
loadPrecisionModels();
|
| 1023 |
} catch (err) {
|
| 1024 |
console.error(err);
|
| 1025 |
+
setPill(els.aiDot, els.aiStatus, "AI model loading failed", "error");
|
| 1026 |
+
showToast("Model loading failed. Refresh the Space and check browser network access.");
|
| 1027 |
}
|
| 1028 |
}
|
| 1029 |
|
| 1030 |
+
async function loadOpenCvFerModel() {
|
| 1031 |
+
await waitForGlobal("ort", 16000);
|
| 1032 |
+
ort.env.wasm.wasmPaths = "https://cdn.jsdelivr.net/npm/onnxruntime-web@1.21.0/dist/";
|
| 1033 |
+
ort.env.wasm.numThreads = Math.max(1, Math.min(4, navigator.hardwareConcurrency || 2));
|
| 1034 |
+
precision.ferSession = await ort.InferenceSession.create(OPENCV_FER_MODEL_URL, { executionProviders: ["wasm"] });
|
| 1035 |
+
updatePrecisionStatus();
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1036 |
}
|
| 1037 |
|
| 1038 |
+
async function loadTransformerEmotionModel() {
|
| 1039 |
+
if (xenovaEnv) {
|
| 1040 |
+
xenovaEnv.allowLocalModels = false;
|
| 1041 |
+
xenovaEnv.useBrowserCache = true;
|
| 1042 |
+
xenovaEnv.backends ??= {};
|
| 1043 |
+
xenovaEnv.backends.onnx ??= {};
|
| 1044 |
+
xenovaEnv.backends.onnx.wasm ??= {};
|
| 1045 |
+
xenovaEnv.backends.onnx.wasm.numThreads = Math.max(1, Math.min(4, navigator.hardwareConcurrency || 2));
|
| 1046 |
+
}
|
| 1047 |
+
precision.emotionPipe = await xenovaPipeline("image-classification", EMOTION_TRANSFORMER_MODEL_ID, { quantized: true });
|
| 1048 |
+
updatePrecisionStatus();
|
| 1049 |
+
}
|
| 1050 |
+
|
| 1051 |
+
async function loadAgeGenderModel() {
|
| 1052 |
+
if (hfEnv) {
|
| 1053 |
+
hfEnv.allowLocalModels = false;
|
| 1054 |
+
hfEnv.useBrowserCache = true;
|
| 1055 |
+
hfEnv.backends ??= {};
|
| 1056 |
+
hfEnv.backends.onnx ??= {};
|
| 1057 |
+
hfEnv.backends.onnx.wasm ??= {};
|
| 1058 |
+
hfEnv.backends.onnx.wasm.numThreads = Math.max(1, Math.min(4, navigator.hardwareConcurrency || 2));
|
| 1059 |
+
}
|
| 1060 |
+
const opts = navigator.gpu ? { device: "webgpu", dtype: "q8" } : { device: "wasm", dtype: "q8" };
|
| 1061 |
+
try {
|
| 1062 |
+
precision.ageModel = await AutoModel.from_pretrained(AGE_GENDER_MODEL_ID, opts);
|
| 1063 |
+
} catch (err) {
|
| 1064 |
+
console.warn("Age/gender preferred backend failed; retrying wasm", err);
|
| 1065 |
+
precision.ageModel = await AutoModel.from_pretrained(AGE_GENDER_MODEL_ID, { device: "wasm", dtype: "q8" });
|
| 1066 |
+
}
|
| 1067 |
+
precision.ageProcessor = await AutoProcessor.from_pretrained(AGE_GENDER_MODEL_ID);
|
| 1068 |
+
updatePrecisionStatus();
|
| 1069 |
+
}
|
| 1070 |
+
|
| 1071 |
+
async function loadPrecisionModels() {
|
| 1072 |
+
if (precision.loading) return;
|
| 1073 |
+
precision.loading = true;
|
| 1074 |
+
updatePrecisionStatus();
|
| 1075 |
+
const tasks = [
|
| 1076 |
+
loadOpenCvFerModel().catch(err => console.warn("OpenCV FER model unavailable", err)),
|
| 1077 |
+
loadTransformerEmotionModel().catch(err => console.warn("Transformer emotion model unavailable", err)),
|
| 1078 |
+
loadAgeGenderModel().catch(err => console.warn("Age/gender transformer unavailable", err)),
|
| 1079 |
];
|
| 1080 |
+
await Promise.allSettled(tasks);
|
| 1081 |
+
precision.loading = false;
|
| 1082 |
+
updatePrecisionStatus();
|
| 1083 |
+
if (precision.ferSession || precision.emotionPipe || precision.ageModel) showToast("Precision models are ready.");
|
| 1084 |
+
}
|
| 1085 |
+
|
| 1086 |
+
function averagePoint(points) {
|
| 1087 |
+
const list = (points || []).map(p => ({ x: p.x, y: p.y })).filter(p => Number.isFinite(p.x) && Number.isFinite(p.y));
|
| 1088 |
+
if (!list.length) return null;
|
| 1089 |
+
return {
|
| 1090 |
+
x: list.reduce((a, p) => a + p.x, 0) / list.length,
|
| 1091 |
+
y: list.reduce((a, p) => a + p.y, 0) / list.length,
|
| 1092 |
+
};
|
| 1093 |
+
}
|
| 1094 |
+
|
| 1095 |
+
function getLandmarkPoints(det) {
|
| 1096 |
+
if (!det.landmarks) return null;
|
| 1097 |
+
const leftEye = averagePoint(det.landmarks.getLeftEye?.() || []);
|
| 1098 |
+
const rightEye = averagePoint(det.landmarks.getRightEye?.() || []);
|
| 1099 |
+
const noseList = det.landmarks.getNose?.() || [];
|
| 1100 |
+
const mouthList = det.landmarks.getMouth?.() || [];
|
| 1101 |
+
const noseTip = noseList.length ? noseList[Math.min(3, noseList.length - 1)] : null;
|
| 1102 |
+
if (!leftEye || !rightEye || !noseTip || !mouthList.length) return null;
|
| 1103 |
+
const eyes = [leftEye, rightEye].sort((a, b) => a.x - b.x);
|
| 1104 |
+
const mouths = mouthList.slice().sort((a, b) => a.x - b.x);
|
| 1105 |
+
const leftMouth = mouths[0];
|
| 1106 |
+
const rightMouth = mouths[mouths.length - 1];
|
| 1107 |
+
return [eyes[0], eyes[1], noseTip, leftMouth, rightMouth];
|
| 1108 |
+
}
|
| 1109 |
+
|
| 1110 |
+
function solveLinearSystem(A, b) {
|
| 1111 |
+
const n = b.length;
|
| 1112 |
+
const M = A.map((row, i) => row.concat([b[i]]));
|
| 1113 |
+
for (let col = 0; col < n; col += 1) {
|
| 1114 |
+
let pivot = col;
|
| 1115 |
+
for (let r = col + 1; r < n; r += 1) if (Math.abs(M[r][col]) > Math.abs(M[pivot][col])) pivot = r;
|
| 1116 |
+
if (Math.abs(M[pivot][col]) < 1e-8) return null;
|
| 1117 |
+
[M[col], M[pivot]] = [M[pivot], M[col]];
|
| 1118 |
+
const div = M[col][col];
|
| 1119 |
+
for (let c = col; c <= n; c += 1) M[col][c] /= div;
|
| 1120 |
+
for (let r = 0; r < n; r += 1) {
|
| 1121 |
+
if (r === col) continue;
|
| 1122 |
+
const factor = M[r][col];
|
| 1123 |
+
for (let c = col; c <= n; c += 1) M[r][c] -= factor * M[col][c];
|
| 1124 |
}
|
| 1125 |
}
|
| 1126 |
+
return M.map(row => row[n]);
|
| 1127 |
+
}
|
| 1128 |
+
|
| 1129 |
+
function estimateAffine(srcPts, dstPts) {
|
| 1130 |
+
const rows = [];
|
| 1131 |
+
const vals = [];
|
| 1132 |
+
for (let i = 0; i < srcPts.length; i += 1) {
|
| 1133 |
+
const sx = srcPts[i].x;
|
| 1134 |
+
const sy = srcPts[i].y;
|
| 1135 |
+
const dx = dstPts[i].x;
|
| 1136 |
+
const dy = dstPts[i].y;
|
| 1137 |
+
rows.push([sx, sy, 1, 0, 0, 0]); vals.push(dx);
|
| 1138 |
+
rows.push([0, 0, 0, sx, sy, 1]); vals.push(dy);
|
| 1139 |
+
}
|
| 1140 |
+
const ATA = Array.from({ length: 6 }, () => Array(6).fill(0));
|
| 1141 |
+
const ATb = Array(6).fill(0);
|
| 1142 |
+
for (let r = 0; r < rows.length; r += 1) {
|
| 1143 |
+
for (let i = 0; i < 6; i += 1) {
|
| 1144 |
+
ATb[i] += rows[r][i] * vals[r];
|
| 1145 |
+
for (let j = 0; j < 6; j += 1) ATA[i][j] += rows[r][i] * rows[r][j];
|
| 1146 |
+
}
|
| 1147 |
+
}
|
| 1148 |
+
return solveLinearSystem(ATA, ATb);
|
| 1149 |
+
}
|
| 1150 |
+
|
| 1151 |
+
function makeAlignedFaceCanvas(det, targetSize = 112) {
|
| 1152 |
+
const points = getLandmarkPoints(det);
|
| 1153 |
+
if (!points) return cropFaceCanvas(det, targetSize, 0.25, "contrast(1.08)");
|
| 1154 |
+
const dst = [
|
| 1155 |
+
{ x: 38.2946, y: 51.6963 },
|
| 1156 |
+
{ x: 73.5318, y: 51.5014 },
|
| 1157 |
+
{ x: 56.0252, y: 71.7366 },
|
| 1158 |
+
{ x: 41.5493, y: 92.3655 },
|
| 1159 |
+
{ x: 70.7299, y: 92.2041 },
|
| 1160 |
+
];
|
| 1161 |
+
const sol = estimateAffine(points, dst);
|
| 1162 |
+
if (!sol) return cropFaceCanvas(det, targetSize, 0.25, "contrast(1.08)");
|
| 1163 |
+
const canvas = document.createElement("canvas");
|
| 1164 |
+
canvas.width = targetSize;
|
| 1165 |
+
canvas.height = targetSize;
|
| 1166 |
+
const c = canvas.getContext("2d", { willReadFrequently: true });
|
| 1167 |
+
c.fillStyle = "#000";
|
| 1168 |
+
c.fillRect(0, 0, targetSize, targetSize);
|
| 1169 |
+
c.setTransform(sol[0], sol[3], sol[1], sol[4], sol[2], sol[5]);
|
| 1170 |
+
c.filter = "contrast(1.08) saturate(0.95)";
|
| 1171 |
+
c.drawImage(els.video, 0, 0);
|
| 1172 |
+
c.setTransform(1, 0, 0, 1, 0, 0);
|
| 1173 |
+
c.filter = "none";
|
| 1174 |
+
return canvas;
|
| 1175 |
}
|
| 1176 |
|
| 1177 |
+
function canvasToNchwTensor(canvas) {
|
| 1178 |
+
const c = canvas.getContext("2d", { willReadFrequently: true });
|
| 1179 |
+
const { data } = c.getImageData(0, 0, canvas.width, canvas.height);
|
| 1180 |
+
const size = canvas.width * canvas.height;
|
| 1181 |
+
const out = new Float32Array(3 * size);
|
| 1182 |
+
for (let i = 0; i < size; i += 1) {
|
| 1183 |
+
const j = i * 4;
|
| 1184 |
+
out[i] = (data[j] / 255 - 0.5) / 0.5;
|
| 1185 |
+
out[size + i] = (data[j + 1] / 255 - 0.5) / 0.5;
|
| 1186 |
+
out[size * 2 + i] = (data[j + 2] / 255 - 0.5) / 0.5;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1187 |
}
|
| 1188 |
+
return out;
|
|
|
|
|
|
|
| 1189 |
}
|
| 1190 |
|
| 1191 |
+
async function runOpenCvFer(det) {
|
| 1192 |
+
if (!precision.ferSession || !window.ort) return null;
|
| 1193 |
+
const canvas = makeAlignedFaceCanvas(det, 112);
|
| 1194 |
+
const input = canvasToNchwTensor(canvas);
|
| 1195 |
+
const tensor = new ort.Tensor("float32", input, [1, 3, 112, 112]);
|
| 1196 |
+
const feeds = {};
|
| 1197 |
+
feeds[precision.ferSession.inputNames[0]] = tensor;
|
| 1198 |
+
const results = await precision.ferSession.run(feeds);
|
| 1199 |
+
const outputName = precision.ferSession.outputNames[0];
|
| 1200 |
+
const logits = Array.from(results[outputName].data).slice(0, 7);
|
| 1201 |
+
const probs = softmax(logits);
|
| 1202 |
+
const scores = blankScores();
|
| 1203 |
+
for (let i = 0; i < probs.length && i < ferLabels.length; i += 1) {
|
| 1204 |
+
const label = ferLabels[i];
|
| 1205 |
+
scores[label] = Math.max(scores[label], probs[i]);
|
| 1206 |
+
}
|
| 1207 |
+
return normalizeScores(scores);
|
| 1208 |
+
}
|
| 1209 |
|
| 1210 |
+
async function runTransformerEmotion(det) {
|
| 1211 |
+
if (!precision.emotionPipe) return null;
|
| 1212 |
+
const variants = [
|
| 1213 |
+
cropFaceCanvas(det, 288, 0.14, "contrast(1.10) saturate(0.95)", false),
|
| 1214 |
+
cropFaceCanvas(det, 288, 0.30, "contrast(1.16) saturate(0.92)", false),
|
| 1215 |
+
cropFaceCanvas(det, 288, 0.08, "contrast(1.18) brightness(1.02)", false),
|
| 1216 |
+
cropFaceCanvas(det, 288, 0.20, "contrast(1.12) saturate(0.94)", true),
|
| 1217 |
+
];
|
| 1218 |
+
const aggregate = blankScores();
|
| 1219 |
+
const urls = [];
|
| 1220 |
+
let count = 0;
|
| 1221 |
try {
|
| 1222 |
+
for (const variant of variants) {
|
| 1223 |
+
const url = await canvasToBlobUrl(variant);
|
| 1224 |
+
urls.push(url);
|
| 1225 |
+
let output;
|
| 1226 |
+
try {
|
| 1227 |
+
output = await precision.emotionPipe(url, { topK: 7 });
|
| 1228 |
+
} catch (_) {
|
| 1229 |
+
output = await precision.emotionPipe(url);
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1230 |
}
|
| 1231 |
+
const scores = normalizeExternalEmotion(output);
|
| 1232 |
+
for (const label of emotionLabels) aggregate[label] += scores[label] || 0;
|
| 1233 |
+
count += 1;
|
| 1234 |
}
|
| 1235 |
+
} finally {
|
| 1236 |
+
for (const url of urls) URL.revokeObjectURL(url);
|
| 1237 |
+
}
|
| 1238 |
+
if (!count) return null;
|
| 1239 |
+
for (const label of emotionLabels) aggregate[label] /= count;
|
| 1240 |
+
return normalizeScores(aggregate);
|
| 1241 |
+
}
|
| 1242 |
|
| 1243 |
+
async function tensorValues(tensor) {
|
| 1244 |
+
if (!tensor) return [];
|
| 1245 |
+
if (tensor.tolist) {
|
| 1246 |
+
const listed = tensor.tolist();
|
| 1247 |
+
return Array.isArray(listed) ? listed.flat(Infinity) : [];
|
| 1248 |
+
}
|
| 1249 |
+
if (tensor.data) return Array.from(tensor.data);
|
| 1250 |
+
return [];
|
| 1251 |
+
}
|
| 1252 |
+
|
| 1253 |
+
async function runAgeGender(det) {
|
| 1254 |
+
if (!precision.ageModel || !precision.ageProcessor || !load_image) return;
|
| 1255 |
+
const canvas = cropFaceCanvas(det, 384, 0.22, "contrast(1.07) saturate(0.96)");
|
| 1256 |
+
let url = null;
|
| 1257 |
+
try {
|
| 1258 |
+
url = await canvasToBlobUrl(canvas);
|
| 1259 |
+
const image = await load_image(url);
|
| 1260 |
+
const inputs = await precision.ageProcessor(image);
|
| 1261 |
+
const output = await precision.ageModel(inputs);
|
| 1262 |
+
const values = await tensorValues(output.logits || output[0]);
|
| 1263 |
+
if (values.length < 2) return;
|
| 1264 |
+
const rawAge = Number(values[0]);
|
| 1265 |
+
const rawGender = Number(values[1]);
|
| 1266 |
+
if (Number.isFinite(rawAge)) {
|
| 1267 |
+
const age = Math.max(0, Math.min(100, Math.round(rawAge)));
|
| 1268 |
+
precision.age = age;
|
| 1269 |
+
precision.ageAt = performance.now();
|
| 1270 |
+
pushAgeSample(age, "precision model", 5);
|
| 1271 |
+
}
|
| 1272 |
+
if (Number.isFinite(rawGender)) {
|
| 1273 |
+
const pFemale = rawGender >= 0 && rawGender <= 1 ? rawGender : sigmoid(rawGender);
|
| 1274 |
+
precision.gender = pFemale >= 0.5 ? "Female" : "Male";
|
| 1275 |
+
precision.genderScore = Math.max(pFemale, 1 - pFemale);
|
| 1276 |
+
precision.genderAt = performance.now();
|
| 1277 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1278 |
} finally {
|
| 1279 |
+
if (url) URL.revokeObjectURL(url);
|
|
|
|
| 1280 |
}
|
| 1281 |
}
|
| 1282 |
|
| 1283 |
+
async function maybeRunPrecision(primary) {
|
|
|
|
| 1284 |
const now = performance.now();
|
| 1285 |
+
if (!primary || precision.busy || (!precision.ferSession && !precision.emotionPipe && !precision.ageModel)) return;
|
| 1286 |
+
if (now - precision.lastRun < detectorConfig.precisionIntervalMs) return;
|
| 1287 |
+
precision.busy = true;
|
| 1288 |
+
precision.lastRun = now;
|
|
|
|
|
|
|
| 1289 |
const started = performance.now();
|
| 1290 |
try {
|
| 1291 |
+
if (precision.ferSession) {
|
| 1292 |
+
const fer = await runOpenCvFer(primary);
|
| 1293 |
+
if (fer) {
|
| 1294 |
+
precision.ferScores = fer;
|
| 1295 |
+
precision.ferAt = performance.now();
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1296 |
}
|
|
|
|
|
|
|
|
|
|
| 1297 |
}
|
| 1298 |
+
if (precision.emotionPipe) {
|
| 1299 |
+
const transformer = await runTransformerEmotion(primary);
|
| 1300 |
+
if (transformer) {
|
| 1301 |
+
precision.transformerScores = transformer;
|
| 1302 |
+
precision.transformerAt = performance.now();
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1303 |
}
|
| 1304 |
}
|
| 1305 |
+
if (precision.ageModel && precision.ageProcessor) await runAgeGender(primary);
|
| 1306 |
} catch (err) {
|
| 1307 |
+
console.warn("Precision inference failed", err);
|
| 1308 |
} finally {
|
| 1309 |
+
precision.latency = performance.now() - started;
|
| 1310 |
+
precision.busy = false;
|
|
|
|
| 1311 |
}
|
| 1312 |
}
|
| 1313 |
|
| 1314 |
function stopStream() {
|
| 1315 |
+
if (stream) for (const track of stream.getTracks()) track.stop();
|
|
|
|
|
|
|
| 1316 |
stream = null;
|
| 1317 |
running = false;
|
| 1318 |
detecting = false;
|
|
|
|
| 1328 |
|
| 1329 |
async function startCamera(facing = currentFacing) {
|
| 1330 |
if (!coreReady) {
|
| 1331 |
+
showToast("Please wait until the AI models finish loading.");
|
| 1332 |
return;
|
| 1333 |
}
|
| 1334 |
|
|
|
|
| 1340 |
audio: false,
|
| 1341 |
video: {
|
| 1342 |
facingMode: { ideal: currentFacing },
|
| 1343 |
+
width: { ideal: 1280 },
|
| 1344 |
height: { ideal: 720 },
|
| 1345 |
}
|
| 1346 |
};
|
|
|
|
| 1373 |
detectLoop();
|
| 1374 |
}
|
| 1375 |
|
| 1376 |
+
async function detectFacesAccurate() {
|
| 1377 |
+
try {
|
| 1378 |
+
const options = new faceapi.SsdMobilenetv1Options({ minConfidence: 0.42 });
|
| 1379 |
+
return await faceapi
|
| 1380 |
+
.detectAllFaces(els.video, options)
|
| 1381 |
+
.withFaceLandmarks(true)
|
| 1382 |
+
.withFaceExpressions()
|
| 1383 |
+
.withAgeAndGender();
|
| 1384 |
+
} catch (err) {
|
| 1385 |
+
console.warn("SSD detector failed; using tiny detector", err);
|
| 1386 |
+
const tiny = new faceapi.TinyFaceDetectorOptions({ inputSize: detectorConfig.inputSize, scoreThreshold: detectorConfig.scoreThreshold });
|
| 1387 |
+
return await faceapi
|
| 1388 |
+
.detectAllFaces(els.video, tiny)
|
| 1389 |
+
.withFaceLandmarks(true)
|
| 1390 |
+
.withFaceExpressions()
|
| 1391 |
+
.withAgeAndGender();
|
| 1392 |
+
}
|
| 1393 |
+
}
|
| 1394 |
+
|
| 1395 |
async function detectLoop() {
|
| 1396 |
if (!running || !coreReady || detecting) return;
|
| 1397 |
detecting = true;
|
| 1398 |
const started = performance.now();
|
| 1399 |
try {
|
| 1400 |
fitCanvas();
|
| 1401 |
+
const raw = await detectFacesAccurate();
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1402 |
const displaySize = { width: els.overlay.width, height: els.overlay.height };
|
| 1403 |
const resized = faceapi.resizeResults(raw, displaySize);
|
| 1404 |
const elapsed = performance.now() - started;
|
|
|
|
| 1407 |
setTimeout(() => {
|
| 1408 |
detecting = false;
|
| 1409 |
requestAnimationFrame(detectLoop);
|
| 1410 |
+
}, detectorConfig.intervalMs);
|
| 1411 |
} catch (err) {
|
| 1412 |
console.error(err);
|
| 1413 |
detecting = false;
|
|
|
|
| 1421 |
currentFacing = currentFacing === "user" ? "environment" : "user";
|
| 1422 |
startCamera(currentFacing);
|
| 1423 |
});
|
|
|
|
| 1424 |
els.stopBtn.addEventListener("click", stopStream);
|
| 1425 |
els.video.addEventListener("loadedmetadata", fitCanvas);
|
| 1426 |
window.addEventListener("resize", fitCanvas);
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1427 |
|
| 1428 |
resetDetails();
|
| 1429 |
renderBars({});
|
| 1430 |
+
setPill(els.precisionDot, els.precisionStatus, "Precision models loading automatically", "loading");
|
| 1431 |
loadCoreModels();
|
| 1432 |
</script>
|
| 1433 |
</body>
|