ambiser
/
human
mirror of https://github.com/vladmandic/human
9
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

add test for face descriptors

pull/91/head
Vladimir Mandic 2021-03-11 18:26:04 -05:00
parent fb10a529ab
commit ca1bc638f8
21 changed files with 298 additions and 136 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored
View File

@ -1,2 +1,2 @@
node_modules
alternative
private

1
TODO.md
View File

@ -2,6 +2,7 @@
- Strong typing
- Automated testing
- Guard against corrupt input
- Improve face embedding
- Build Face embedding database
- Dynamic sample processing

1
config.js
View File

@ -85,6 +85,7 @@ export default {
scoreThreshold: 0.2, // threshold for deciding when to remove boxes based on score
// in non-maximum suppression,
// this is applied on detection objects only and before minConfidence
return: true, // return extracted face as tensor
},
mesh: {

32
demo/embedding.html Normal file
View File

@ -0,0 +1,32 @@
<!DOCTYPE html>
<html lang="en">
<head>
<title>Human</title>
<meta charset="utf-8">
<meta http-equiv="content-type">
<meta content="text/html">
<meta name="description" content="Human: 3D Face Detection, Body Pose, Hand & Finger Tracking, Iris Tracking, Age & Gender Prediction, Emotion Prediction & Gesture Recognition; Author: Vladimir Mandic <https://github.com/vladmandic>">
<meta name="viewport" content="width=device-width, initial-scale=0.6, minimum-scale=0.3, maximum-scale=3.0, shrink-to-fit=yes, user-scalable=yes">
<meta name="theme-color" content="#000000"/>
<meta name="application-name" content="Human">
<meta name="msapplication-tooltip" content="Human: AI-powered 3D Human Detection">
<link rel="manifest" href="./manifest.webmanifest">
<link rel="shortcut icon" href="../favicon.ico" type="image/x-icon">
<link rel="apple-touch-icon" href="../assets/icon.png">
<script src="./embedding.js" type="module"></script>
<style>
@font-face { font-family: 'Lato'; font-display: swap; font-style: normal; font-weight: 100; src: local('Lato'), url('../assets/lato-light.woff2') }
html { font-family: 'Lato', 'Segoe UI'; font-size: 20px; font-variant: small-caps; }
body { margin: 0; background: black; color: white; overflow-x: hidden; scrollbar-width: none; }
body::-webkit-scrollbar { display: none; }
img { object-fit: contain; }
.face { width: 200px; height: 200px; }
</style>
</head>
<body>
<br>Sample Images:
<div id="images"></div>
<br>Extracted Faces - click on a face to sort by simmilarity:<br>
<div id="faces"></div>
</body>
</html>

94
demo/embedding.js Normal file
View File

@ -0,0 +1,94 @@
import Human from '../dist/human.esm.js';
const userConfig = {
backend: 'wasm',
async: false,
warmup: 'none',
debug: true,
filter: false,
videoOptimized: false,
face: {
enabled: true,
detector: { rotation: true },
mesh: { enabled: true },
embedding: { enabled: true, modelPath: '../models/mobilefacenet.json' },
iris: { enabled: false },
age: { enabled: false },
gender: { enabled: false },
emotion: { enabled: false },
},
hand: { enabled: false },
gesture: { enabled: false },
body: { enabled: false },
};
const human = new Human(userConfig);
const samples = ['../assets/sample-me.jpg', '../assets/sample6.jpg', '../assets/sample1.jpg', '../assets/sample4.jpg', '../assets/sample5.jpg', '../assets/sample3.jpg', '../assets/sample2.jpg',
'../private/me (1).jpg', '../private/me (2).jpg', '../private/me (3).jpg', '../private/me (4).jpg', '../private/me (5).jpg', '../private/me (6).jpg', '../private/me (7).jpg', '../private/me (8).jpg',
'../private/me (9).jpg', '../private/me (10).jpg', '../private/me (11).jpg', '../private/me (12).jpg', '../private/me (13).jpg'];
const all = [];
function log(...msg) {
const dt = new Date();
const ts = `${dt.getHours().toString().padStart(2, '0')}:${dt.getMinutes().toString().padStart(2, '0')}:${dt.getSeconds().toString().padStart(2, '0')}.${dt.getMilliseconds().toString().padStart(3, '0')}`;
// eslint-disable-next-line no-console
console.log(ts, ...msg);
}
async function analyze(face) {
log('Face:', face);
const canvases = document.getElementsByClassName('face');
for (const canvas of canvases) {
const res = human.simmilarity(face.embedding, all[canvas.tag.sample][canvas.tag.face].embedding);
canvas.title = res;
await human.tf.browser.toPixels(all[canvas.tag.sample][canvas.tag.face].tensor, canvas);
const ctx = canvas.getContext('2d');
ctx.font = 'small-caps 1rem "Lato"';
ctx.fillStyle = 'rgba(0, 0, 0, 1)';
ctx.fillText(`${(100 * res).toFixed(1)}%`, 3, 19);
ctx.fillStyle = 'rgba(255, 255, 255, 1)';
ctx.fillText(`${(100 * res).toFixed(1)}%`, 4, 20);
}
const sorted = document.getElementById('faces');
[...sorted.children]
.sort((a, b) => parseFloat(b.title) - parseFloat(a.title))
.forEach((canvas) => sorted.appendChild(canvas));
}
async function faces(index, res) {
all[index] = res.face;
for (const i in res.face) {
// log(res.face[i]);
const canvas = document.createElement('canvas');
canvas.tag = { sample: index, face: i };
canvas.width = 200;
canvas.height = 200;
canvas.className = 'face';
canvas.addEventListener('click', (evt) => {
log('Select:', evt.target.tag.sample, evt.target.tag.face);
analyze(all[evt.target.tag.sample][evt.target.tag.face]);
});
human.tf.browser.toPixels(res.face[i].tensor, canvas);
document.getElementById('faces').appendChild(canvas);
}
}
async function add(index) {
log('Add:', samples[index]);
return new Promise((resolve) => {
const img = new Image(100, 100);
img.onload = () => {
human.detect(img).then((res) => faces(index, res));
document.getElementById('images').appendChild(img);
resolve(true);
};
img.title = samples[index];
img.src = samples[index];
});
}
async function main() {
await human.load();
for (const i in samples) await add(i);
}
window.onload = main;

35
dist/demo-browser-index.js vendored
View File

@ -79445,32 +79445,39 @@ function simmilarity(embedding1, embedding22, order = 2) {
return 0;
if ((embedding1 == null ? void 0 : embedding1.length) !== (embedding22 == null ? void 0 : embedding22.length))
return 0;
const distance = 50 * embedding1.map((val, i) => val - embedding22[i]).reduce((dist, diff) => dist + diff ** order, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - (isNaN(distance) ? 1 : distance))) / 1e3;
console.log(distance, res);
const distance = embedding1.map((val, i) => Math.abs(embedding1[i] - embedding22[i]) ** order).reduce((sum6, now3) => sum6 + now3, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - 20 * distance)) / 1e3;
return res;
}
async function predict4(image3, config3) {
async function predict4(input2, config3) {
if (!model5)
return null;
return new Promise(async (resolve) => {
const resize = image.resizeBilinear(image3, [model5.inputs[0].shape[2], model5.inputs[0].shape[1]], false);
const norm2 = resize.sub(0.5);
resize.dispose();
const image3 = tidy(() => {
const data3 = image.resizeBilinear(input2, [model5.inputs[0].shape[2], model5.inputs[0].shape[1]], false);
const norm2 = data3.sub(data3.mean());
return norm2;
});
let data2 = [];
if (config3.face.embedding.enabled) {
if (!config3.profile) {
const res = await model5.predict({img_inputs: norm2});
data2 = [...res.dataSync()];
const res = await model5.predict({img_inputs: image3});
const scaled = tidy(() => {
const l23 = res.norm("euclidean");
const scale2 = res.div(l23);
return scale2;
});
data2 = [...scaled.dataSync()];
dispose(scaled);
dispose(res);
} else {
const profileData = await profile(() => model5.predict({img_inputs: norm2}));
const profileData = await profile(() => model5.predict({img_inputs: image3}));
data2 = [...profileData.result.dataSync()];
profileData.result.dispose();
run("emotion", profileData);
}
}
norm2.dispose();
image3.dispose();
resolve(data2);
});
}
@ -99022,7 +99029,8 @@ var config_default = {
skipInitial: false,
minConfidence: 0.2,
iouThreshold: 0.1,
scoreThreshold: 0.2
scoreThreshold: 0.2,
return: true
},
mesh: {
enabled: true,
@ -100505,7 +100513,8 @@ var Human = class {
emotion: emotionRes,
embedding: embeddingRes,
iris: irisSize !== 0 ? Math.trunc(irisSize) / 100 : 0,
angle
angle,
tensor: this.config.face.detector.return ? face4.image.squeeze() : null
});
(_f = face4.image) == null ? void 0 : _f.dispose();
__privateGet(this, _analyze).call(this, "End Face");

4
dist/demo-browser-index.js.map vendored
View File

File diff suppressed because one or more lines are too long

35
dist/human.esm-nobundle.js vendored
View File

@ -4357,32 +4357,39 @@ function simmilarity(embedding1, embedding2, order = 2) {
return 0;
if ((embedding1 == null ? void 0 : embedding1.length) !== (embedding2 == null ? void 0 : embedding2.length))
return 0;
const distance = 50 * embedding1.map((val, i) => val - embedding2[i]).reduce((dist, diff) => dist + diff ** order, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - (isNaN(distance) ? 1 : distance))) / 1e3;
console.log(distance, res);
const distance = embedding1.map((val, i) => Math.abs(embedding1[i] - embedding2[i]) ** order).reduce((sum, now2) => sum + now2, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - 20 * distance)) / 1e3;
return res;
}
async function predict4(image13, config3) {
async function predict4(input, config3) {
if (!model4)
return null;
return new Promise(async (resolve) => {
const resize = tf8.image.resizeBilinear(image13, [model4.inputs[0].shape[2], model4.inputs[0].shape[1]], false);
const norm = resize.sub(0.5);
resize.dispose();
const image13 = tf8.tidy(() => {
const data4 = tf8.image.resizeBilinear(input, [model4.inputs[0].shape[2], model4.inputs[0].shape[1]], false);
const norm = data4.sub(data4.mean());
return norm;
});
let data3 = [];
if (config3.face.embedding.enabled) {
if (!config3.profile) {
const res = await model4.predict({img_inputs: norm});
data3 = [...res.dataSync()];
const res = await model4.predict({img_inputs: image13});
const scaled = tf8.tidy(() => {
const l2 = res.norm("euclidean");
const scale = res.div(l2);
return scale;
});
data3 = [...scaled.dataSync()];
tf8.dispose(scaled);
tf8.dispose(res);
} else {
const profileData = await tf8.profile(() => model4.predict({img_inputs: norm}));
const profileData = await tf8.profile(() => model4.predict({img_inputs: image13}));
data3 = [...profileData.result.dataSync()];
profileData.result.dispose();
run("emotion", profileData);
}
}
norm.dispose();
image13.dispose();
resolve(data3);
});
}
@ -23949,7 +23956,8 @@ var config_default = {
skipInitial: false,
minConfidence: 0.2,
iouThreshold: 0.1,
scoreThreshold: 0.2
scoreThreshold: 0.2,
return: true
},
mesh: {
enabled: true,
@ -25432,7 +25440,8 @@ var Human = class {
emotion: emotionRes,
embedding: embeddingRes,
iris: irisSize !== 0 ? Math.trunc(irisSize) / 100 : 0,
angle
angle,
tensor: this.config.face.detector.return ? face4.image.squeeze() : null
});
(_f = face4.image) == null ? void 0 : _f.dispose();
__privateGet(this, _analyze).call(this, "End Face");

4
dist/human.esm-nobundle.js.map vendored
View File

File diff suppressed because one or more lines are too long

35
dist/human.esm.js vendored
View File

@ -79445,32 +79445,39 @@ function simmilarity(embedding1, embedding22, order = 2) {
return 0;
if ((embedding1 == null ? void 0 : embedding1.length) !== (embedding22 == null ? void 0 : embedding22.length))
return 0;
const distance = 50 * embedding1.map((val, i) => val - embedding22[i]).reduce((dist, diff) => dist + diff ** order, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - (isNaN(distance) ? 1 : distance))) / 1e3;
console.log(distance, res);
const distance = embedding1.map((val, i) => Math.abs(embedding1[i] - embedding22[i]) ** order).reduce((sum6, now3) => sum6 + now3, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - 20 * distance)) / 1e3;
return res;
}
async function predict4(image3, config3) {
async function predict4(input2, config3) {
if (!model5)
return null;
return new Promise(async (resolve) => {
const resize = image.resizeBilinear(image3, [model5.inputs[0].shape[2], model5.inputs[0].shape[1]], false);
const norm2 = resize.sub(0.5);
resize.dispose();
const image3 = tidy(() => {
const data3 = image.resizeBilinear(input2, [model5.inputs[0].shape[2], model5.inputs[0].shape[1]], false);
const norm2 = data3.sub(data3.mean());
return norm2;
});
let data2 = [];
if (config3.face.embedding.enabled) {
if (!config3.profile) {
const res = await model5.predict({img_inputs: norm2});
data2 = [...res.dataSync()];
const res = await model5.predict({img_inputs: image3});
const scaled = tidy(() => {
const l23 = res.norm("euclidean");
const scale2 = res.div(l23);
return scale2;
});
data2 = [...scaled.dataSync()];
dispose(scaled);
dispose(res);
} else {
const profileData = await profile(() => model5.predict({img_inputs: norm2}));
const profileData = await profile(() => model5.predict({img_inputs: image3}));
data2 = [...profileData.result.dataSync()];
profileData.result.dispose();
run("emotion", profileData);
}
}
norm2.dispose();
image3.dispose();
resolve(data2);
});
}
@ -99022,7 +99029,8 @@ var config_default = {
skipInitial: false,
minConfidence: 0.2,
iouThreshold: 0.1,
scoreThreshold: 0.2
scoreThreshold: 0.2,
return: true
},
mesh: {
enabled: true,
@ -100505,7 +100513,8 @@ var Human = class {
emotion: emotionRes,
embedding: embeddingRes,
iris: irisSize !== 0 ? Math.trunc(irisSize) / 100 : 0,
angle
angle,
tensor: this.config.face.detector.return ? face4.image.squeeze() : null
});
(_f = face4.image) == null ? void 0 : _f.dispose();
__privateGet(this, _analyze).call(this, "End Face");

4
dist/human.esm.js.map vendored
View File

File diff suppressed because one or more lines are too long

35
dist/human.js vendored
View File

@ -79452,32 +79452,39 @@ return a / b;`;
return 0;
if ((embedding1 == null ? void 0 : embedding1.length) !== (embedding22 == null ? void 0 : embedding22.length))
return 0;
const distance = 50 * embedding1.map((val, i) => val - embedding22[i]).reduce((dist, diff) => dist + diff ** order, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - (isNaN(distance) ? 1 : distance))) / 1e3;
console.log(distance, res);
const distance = embedding1.map((val, i) => Math.abs(embedding1[i] - embedding22[i]) ** order).reduce((sum6, now3) => sum6 + now3, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - 20 * distance)) / 1e3;
return res;
}
async function predict4(image3, config3) {
async function predict4(input2, config3) {
if (!model5)
return null;
return new Promise(async (resolve) => {
const resize = image.resizeBilinear(image3, [model5.inputs[0].shape[2], model5.inputs[0].shape[1]], false);
const norm2 = resize.sub(0.5);
resize.dispose();
const image3 = tidy(() => {
const data3 = image.resizeBilinear(input2, [model5.inputs[0].shape[2], model5.inputs[0].shape[1]], false);
const norm2 = data3.sub(data3.mean());
return norm2;
});
let data2 = [];
if (config3.face.embedding.enabled) {
if (!config3.profile) {
const res = await model5.predict({img_inputs: norm2});
data2 = [...res.dataSync()];
const res = await model5.predict({img_inputs: image3});
const scaled = tidy(() => {
const l23 = res.norm("euclidean");
const scale2 = res.div(l23);
return scale2;
});
data2 = [...scaled.dataSync()];
dispose(scaled);
dispose(res);
} else {
const profileData = await profile(() => model5.predict({img_inputs: norm2}));
const profileData = await profile(() => model5.predict({img_inputs: image3}));
data2 = [...profileData.result.dataSync()];
profileData.result.dispose();
run("emotion", profileData);
}
}
norm2.dispose();
image3.dispose();
resolve(data2);
});
}
@ -99029,7 +99036,8 @@ return a / b;`;
skipInitial: false,
minConfidence: 0.2,
iouThreshold: 0.1,
scoreThreshold: 0.2
scoreThreshold: 0.2,
return: true
},
mesh: {
enabled: true,
@ -100512,7 +100520,8 @@ lBhEMohlFerLlBjEMohMVTEARDKCITsAk2AEgAAAkAAAAAAAAAAAAAAAAAAAAAAAASAAAAAAAAD/
emotion: emotionRes,
embedding: embeddingRes,
iris: irisSize !== 0 ? Math.trunc(irisSize) / 100 : 0,
angle
angle,
tensor: this.config.face.detector.return ? face4.image.squeeze() : null
});
(_f = face4.image) == null ? void 0 : _f.dispose();
__privateGet(this, _analyze).call(this, "End Face");

4
dist/human.js.map vendored
View File

File diff suppressed because one or more lines are too long

35
dist/human.node-gpu.js vendored
View File

@ -4337,32 +4337,39 @@ function simmilarity(embedding1, embedding2, order = 2) {
return 0;
if ((embedding1 == null ? void 0 : embedding1.length) !== (embedding2 == null ? void 0 : embedding2.length))
return 0;
const distance = 50 * embedding1.map((val, i) => val - embedding2[i]).reduce((dist, diff) => dist + diff ** order, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - (isNaN(distance) ? 1 : distance))) / 1e3;
console.log(distance, res);
const distance = embedding1.map((val, i) => Math.abs(embedding1[i] - embedding2[i]) ** order).reduce((sum, now2) => sum + now2, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - 20 * distance)) / 1e3;
return res;
}
async function predict4(image13, config3) {
async function predict4(input, config3) {
if (!model4)
return null;
return new Promise(async (resolve) => {
const resize = tf8.image.resizeBilinear(image13, [model4.inputs[0].shape[2], model4.inputs[0].shape[1]], false);
const norm = resize.sub(0.5);
resize.dispose();
const image13 = tf8.tidy(() => {
const data3 = tf8.image.resizeBilinear(input, [model4.inputs[0].shape[2], model4.inputs[0].shape[1]], false);
const norm = data3.sub(data3.mean());
return norm;
});
let data2 = [];
if (config3.face.embedding.enabled) {
if (!config3.profile) {
const res = await model4.predict({img_inputs: norm});
data2 = [...res.dataSync()];
const res = await model4.predict({img_inputs: image13});
const scaled = tf8.tidy(() => {
const l2 = res.norm("euclidean");
const scale = res.div(l2);
return scale;
});
data2 = [...scaled.dataSync()];
tf8.dispose(scaled);
tf8.dispose(res);
} else {
const profileData = await tf8.profile(() => model4.predict({img_inputs: norm}));
const profileData = await tf8.profile(() => model4.predict({img_inputs: image13}));
data2 = [...profileData.result.dataSync()];
profileData.result.dispose();
run("emotion", profileData);
}
}
norm.dispose();
image13.dispose();
resolve(data2);
});
}
@ -23929,7 +23936,8 @@ var config_default = {
skipInitial: false,
minConfidence: 0.2,
iouThreshold: 0.1,
scoreThreshold: 0.2
scoreThreshold: 0.2,
return: true
},
mesh: {
enabled: true,
@ -25412,7 +25420,8 @@ var Human = class {
emotion: emotionRes,
embedding: embeddingRes,
iris: irisSize !== 0 ? Math.trunc(irisSize) / 100 : 0,
angle
angle,
tensor: this.config.face.detector.return ? face4.image.squeeze() : null
});
(_f = face4.image) == null ? void 0 : _f.dispose();
__privateGet(this, _analyze).call(this, "End Face");

4
dist/human.node-gpu.js.map vendored
View File

File diff suppressed because one or more lines are too long

35
dist/human.node.js vendored
View File

@ -4337,32 +4337,39 @@ function simmilarity(embedding1, embedding2, order = 2) {
return 0;
if ((embedding1 == null ? void 0 : embedding1.length) !== (embedding2 == null ? void 0 : embedding2.length))
return 0;
const distance = 50 * embedding1.map((val, i) => val - embedding2[i]).reduce((dist, diff) => dist + diff ** order, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - (isNaN(distance) ? 1 : distance))) / 1e3;
console.log(distance, res);
const distance = embedding1.map((val, i) => Math.abs(embedding1[i] - embedding2[i]) ** order).reduce((sum, now2) => sum + now2, 0) ** (1 / order);
const res = Math.trunc(1e3 * (1 - 20 * distance)) / 1e3;
return res;
}
async function predict4(image13, config3) {
async function predict4(input, config3) {
if (!model4)
return null;
return new Promise(async (resolve) => {
const resize = tf8.image.resizeBilinear(image13, [model4.inputs[0].shape[2], model4.inputs[0].shape[1]], false);
const norm = resize.sub(0.5);
resize.dispose();
const image13 = tf8.tidy(() => {
const data3 = tf8.image.resizeBilinear(input, [model4.inputs[0].shape[2], model4.inputs[0].shape[1]], false);
const norm = data3.sub(data3.mean());
return norm;
});
let data2 = [];
if (config3.face.embedding.enabled) {
if (!config3.profile) {
const res = await model4.predict({img_inputs: norm});
data2 = [...res.dataSync()];
const res = await model4.predict({img_inputs: image13});
const scaled = tf8.tidy(() => {
const l2 = res.norm("euclidean");
const scale = res.div(l2);
return scale;
});
data2 = [...scaled.dataSync()];
tf8.dispose(scaled);
tf8.dispose(res);
} else {
const profileData = await tf8.profile(() => model4.predict({img_inputs: norm}));
const profileData = await tf8.profile(() => model4.predict({img_inputs: image13}));
data2 = [...profileData.result.dataSync()];
profileData.result.dispose();
run("emotion", profileData);
}
}
norm.dispose();
image13.dispose();
resolve(data2);
});
}
@ -23929,7 +23936,8 @@ var config_default = {
skipInitial: false,
minConfidence: 0.2,
iouThreshold: 0.1,
scoreThreshold: 0.2
scoreThreshold: 0.2,
return: true
},
mesh: {
enabled: true,
@ -25412,7 +25420,8 @@ var Human = class {
emotion: emotionRes,
embedding: embeddingRes,
iris: irisSize !== 0 ? Math.trunc(irisSize) / 100 : 0,
angle
angle,
tensor: this.config.face.detector.return ? face4.image.squeeze() : null
});
(_f = face4.image) == null ? void 0 : _f.dispose();
__privateGet(this, _analyze).call(this, "End Face");

4
dist/human.node.js.map vendored
View File

File diff suppressed because one or more lines are too long

61
src/embedding/embedding.ts
View File

@ -6,13 +6,6 @@ import * as profile from '../profile';
// modified: https://github.com/sirius-ai/MobileFaceNet_TF/issues/46
// download: https://github.com/sirius-ai/MobileFaceNet_TF/files/3551493/FaceMobileNet192_train_false.zip
/* WiP
- Should input box be tightly cropped?
- What is the best input range? (adjust distance scale accordingly)
- How to best normalize output
*/
let model;
export async function load(config) {
@ -29,58 +22,44 @@ export function simmilarity(embedding1, embedding2, order = 2) {
if (embedding1?.length !== embedding2?.length) return 0;
// general minkowski distance
// euclidean distance is limited case where order is 2
const distance = 50.0 * ((embedding1.map((val, i) => (val - embedding2[i])).reduce((dist, diff) => dist + (diff ** order), 0) ** (1 / order)));
const res = (Math.trunc(1000 * (1 - (isNaN(distance) ? 1 : distance))) / 1000);
console.log(distance, res);
const distance = embedding1
.map((val, i) => (Math.abs(embedding1[i] - embedding2[i]) ** order)) // distance squared
.reduce((sum, now) => (sum + now), 0) // sum all distances
** (1 / order); // get root of
const res = Math.trunc(1000 * (1 - (20 * distance))) / 1000;
return res;
}
export async function predict(image, config) {
export async function predict(input, config) {
if (!model) return null;
return new Promise(async (resolve) => {
const resize = tf.image.resizeBilinear(image, [model.inputs[0].shape[2], model.inputs[0].shape[1]], false); // input is already normalized to 0..1
// optionally do a tight box crop
/*
const box = [[0, 0.2, 0.9, 0.8]]; // top, left, bottom, right
const resize = tf.image.cropAndResize(image, box, [0], [model.inputs[0].shape[2], model.inputs[0].shape[1]]);
*/
// debug visualize box
// const canvas = document.getElementById('compare-canvas');
// await tf.browser.toPixels(resize.squeeze(), canvas);
const norm = resize.sub(0.5);
// optionally normalizes with mean value being at point 0, better than fixed range -0.5..0.5
/*
const mean = resize.mean();
const norm = resize.sub(mean);
*/
resize.dispose();
const image = tf.tidy(() => {
const data = tf.image.resizeBilinear(input, [model.inputs[0].shape[2], model.inputs[0].shape[1]], false); // input is already normalized to 0..1
// const box = [[0.05, 0.10, 0.85, 0.90]]; // top, left, bottom, right
// const crop = tf.image.cropAndResize(data, box, [0], [model.inputs[0].shape[2], model.inputs[0].shape[1]]); // optionally do a tight box crop
const norm = data.sub(data.mean()); // trick to normalize around image mean value
return norm;
});
let data: Array<[]> = [];
if (config.face.embedding.enabled) {
if (!config.profile) {
const res = await model.predict({ img_inputs: norm });
/*
const res = await model.predict({ img_inputs: image });
const scaled = tf.tidy(() => {
// run l2 normalization on output
const sqr = res.square();
const sum = sqr.sum();
const sqrt = sum.sqrt();
const l2 = res.div(sqrt);
// scale outputs
const range = l2.max().sub(l2.min());
const scale = l2.mul(2).div(range);
const l2 = res.norm('euclidean');
const scale = res.div(l2);
return scale;
});
*/
data = [...res.dataSync()]; // convert object array to standard array
data = [...scaled.dataSync()]; // convert object array to standard array
tf.dispose(scaled);
tf.dispose(res);
} else {
const profileData = await tf.profile(() => model.predict({ img_inputs: norm }));
const profileData = await tf.profile(() => model.predict({ img_inputs: image }));
data = [...profileData.result.dataSync()];
profileData.result.dispose();
profile.run('emotion', profileData);
}
}
norm.dispose();
image.dispose();
resolve(data);
});
}

5
src/human.ts
View File

@ -303,7 +303,8 @@ class Human {
emotion: string,
embedding: any,
iris: number,
angle: any
angle: any,
tensor: any,
}> = [];
this.state = 'run:face';
@ -402,7 +403,7 @@ class Human {
embedding: embeddingRes,
iris: (irisSize !== 0) ? Math.trunc(irisSize) / 100 : 0,
angle,
// image: face.image.toInt().squeeze(),
tensor: this.config.face.detector.return ? face.image.squeeze() : null,
});
// dont need face anymore

2
types/embedding/embedding.d.ts vendored
View File

@ -1,3 +1,3 @@
export declare function load(config: any): Promise<any>;
export declare function simmilarity(embedding1: any, embedding2: any, order?: number): number;
export declare function predict(image: any, config: any): Promise<unknown>;
export declare function predict(input: any, config: any): Promise<unknown>;

2
wiki

@ -1 +1 @@
Subproject commit 69294f7a0a99bd996286f8f5bb655c7ea8bfc10d
Subproject commit fa7ac1f695547aa0fd25845e6cac7ed5ee0adcae