forked from webmachinelearning/webnn-samples
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.js
323 lines (305 loc) · 11.6 KB
/
main.js
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
'use strict';
import {FaceLandmarkNhwc} from './face_landmark_nhwc.js';
import {FaceLandmarkNchw} from './face_landmark_nchw.js';
import {SsdMobilenetV2FaceNhwc} from './ssd_mobilenetv2_face_nhwc.js';
import {SsdMobilenetV2FaceNchw} from './ssd_mobilenetv2_face_nchw.js';
import * as ui from '../common/ui.js';
import * as utils from '../common/utils.js';
import * as SsdDecoder from '../common/libs/ssdDecoder.js';
import * as FaceLandmark from './libs/face_landmark_utils.js';
const imgElement = document.getElementById('feedElement');
imgElement.src = './images/test.jpg';
const camElement = document.getElementById('feedMediaElement');
let fdModelName = '';
const fldModelName = 'facelandmark';
let layout = 'nhwc';
let fdInstanceType = fdModelName + layout;
let fldInstanceType = fldModelName + layout;
let rafReq;
let isFirstTimeLoad = true;
let inputType = 'image';
let fdInstance = null;
let fdInputOptions;
let fldInstance = null;
let fldInputOptions;
let stream = null;
let loadTime = 0;
let buildTime = 0;
let computeTime = 0;
let fdOutputs;
let fldOutputs;
let deviceType = '';
let lastdeviceType = '';
let backend = '';
let lastBackend = '';
const disabledSelectors = ['#tabs > li', '.btn'];
$(document).ready(async () => {
$('.icdisplay').hide();
if (await utils.isWebNN()) {
$('#webnn_cpu').click();
} else {
$('#polyfill_cpu').click();
}
});
$('#backendBtns .btn').on('change', async (e) => {
if (inputType === 'camera') utils.stopCameraStream(rafReq, stream);
if ($(e.target).attr('id').indexOf('cpu') != -1) {
layout = 'nhwc';
} else if (($(e.target).attr('id').indexOf('gpu') != -1)) {
layout = 'nchw';
} else {
throw new Error('Unknown backend');
}
await main();
});
$('#fdModelBtns .btn').on('change', async (e) => {
fdModelName = $(e.target).attr('id');
if (inputType === 'camera') utils.stopCameraStream(rafReq, stream);
await main();
});
// $('#layoutBtns .btn').on('change', async (e) => {
// layout = $(e.target).attr('id');
// if (inputType === 'camera') utils.stopCameraStream(rafReq, stream);
// await main();
// });
// Click trigger to do inference with <img> element
$('#img').click(async () => {
if (inputType === 'camera') utils.stopCameraStream(rafReq, stream);
inputType = 'image';
$('.shoulddisplay').hide();
await main();
});
$('#imageFile').change((e) => {
const files = e.target.files;
if (files.length > 0) {
$('#feedElement').removeAttr('height');
$('#feedElement').removeAttr('width');
imgElement.src = URL.createObjectURL(files[0]);
}
});
$('#feedElement').on('load', async () => {
await main();
});
// Click trigger to do inference with <video> media element
$('#cam').click(async () => {
inputType = 'camera';
$('.shoulddisplay').hide();
await main();
});
/**
* This method is used to render live camera tab.
*/
async function renderCamStream() {
if (!stream.active) return;
// If the video element's readyState is 0, the video's width and height are 0.
// So check the readState here to make sure it is greater than 0.
if (camElement.readyState === 0) {
rafReq = requestAnimationFrame(renderCamStream);
return;
}
const inputCanvas = utils.getVideoFrame(camElement);
console.log('- Computing... ');
const [totalComputeTime, strokedRects, keyPoints] =
await predict(camElement);
console.log(` done in ${totalComputeTime} ms.`);
computeTime = totalComputeTime;
showPerfResult();
await drawOutput(inputCanvas, strokedRects, keyPoints);
$('#fps').text(`${(1000/totalComputeTime).toFixed(0)} FPS`);
rafReq = requestAnimationFrame(renderCamStream);
}
async function predict(inputElement) {
const fdInputBuffer = utils.getInputTensor(inputElement, fdInputOptions);
let totalComputeTime = 0;
let start = performance.now();
const results = await fdInstance.compute(fdInputBuffer, fdOutputs);
totalComputeTime += performance.now() - start;
fdOutputs = results.outputs;
const strokedRects = [];
const keyPoints = [];
const height = inputElement.naturalHeight || inputElement.height;
const width = inputElement.naturalWidth || inputElement.width;
const fdOutputArrary = [];
for (const output of Object.entries(fdOutputs)) {
fdOutputArrary.push(output[1]);
}
const fdSsdOutputs = SsdDecoder.processSsdOutputTensor(
fdOutputArrary, fdInputOptions, fdInstance.outputsInfo);
const anchors = SsdDecoder.generateAnchors({});
SsdDecoder.decodeOutputBoxTensor({}, fdSsdOutputs.outputBoxTensor, anchors);
let [totalDetections, boxesList, scoresList] = SsdDecoder.nonMaxSuppression(
{numClasses: 2},
fdSsdOutputs.outputBoxTensor,
fdSsdOutputs.outputClassScoresTensor);
boxesList = SsdDecoder.cropSsdBox(
inputElement, totalDetections, boxesList, fdInputOptions.margin);
for (let i = 0; i < totalDetections; ++i) {
let [ymin, xmin, ymax, xmax] = boxesList[i];
ymin = Math.max(0, ymin) * height;
xmin = Math.max(0, xmin) * width;
ymax = Math.min(1, ymax) * height;
xmax = Math.min(1, xmax) * width;
const prob = 1 / (1 + Math.exp(-scoresList[i]));
const rect = [xmin, ymin, xmax - xmin, ymax - ymin, prob];
strokedRects.push(rect);
const drawOptions= {
sx: xmin,
sy: ymin,
sWidth: rect[2],
sHeight: rect[3],
dWidth: 128,
dHeight: 128,
};
fldInputOptions.drawOptions = drawOptions;
const fldInputBuffer = utils.getInputTensor(inputElement, fldInputOptions);
start = performance.now();
const results = await fldInstance.compute(fldInputBuffer, fldOutputs);
totalComputeTime += performance.now() - start;
fldOutputs = results.outputs;
keyPoints.push(fldOutputs.output.slice());
}
return [totalComputeTime.toFixed(2), strokedRects, keyPoints];
}
async function drawOutput(inputElement, strokedRects, keyPoints) {
const outputElement = document.getElementById('outputCanvas');
$('#inferenceresult').show();
const texts = strokedRects.map((r) => r[4].toFixed(2));
SsdDecoder.drawFaceRectangles(
inputElement, outputElement, strokedRects, texts);
FaceLandmark.drawKeyPoints(
inputElement, outputElement, keyPoints, strokedRects);
}
function showPerfResult(medianComputeTime = undefined) {
$('#loadTime').html(`${loadTime} ms`);
$('#buildTime').html(`${buildTime} ms`);
if (medianComputeTime !== undefined) {
$('#computeLabel').html('Median inference time:');
$('#computeTime').html(`${medianComputeTime} ms`);
} else {
$('#computeLabel').html('Inference time:');
$('#computeTime').html(`${computeTime} ms`);
}
}
function constructNetObject(type) {
const netObject = {
'ssdmobilenetv2facenchw': new SsdMobilenetV2FaceNchw(),
'ssdmobilenetv2facenhwc': new SsdMobilenetV2FaceNhwc(),
'facelandmarknchw': new FaceLandmarkNchw(),
'facelandmarknhwc': new FaceLandmarkNhwc(),
};
return netObject[type];
}
async function main() {
try {
if (fdModelName === '') return;
[backend, deviceType] =
$('input[name="backend"]:checked').attr('id').split('_');
ui.handleClick(disabledSelectors, true);
if (isFirstTimeLoad) $('#hint').hide();
const [numRuns, powerPreference, numThreads] = utils.getUrlParams();
let start;
// Only do load() and build() when model first time loads,
// there's new model choosed, backend changed or device changed
if (isFirstTimeLoad || fdInstanceType !== fdModelName + layout ||
lastdeviceType != deviceType || lastBackend != backend) {
if (lastdeviceType != deviceType || lastBackend != backend) {
// Set backend and device
await utils.setBackend(backend, deviceType);
lastdeviceType = lastdeviceType != deviceType ?
deviceType : lastdeviceType;
lastBackend = lastBackend != backend ? backend : lastBackend;
}
if (fldInstance !== null) {
// Call dispose() to and avoid memory leak
fldInstance.dispose();
}
if (fdInstance !== null) {
// Call dispose() to and avoid memory leak
fdInstance.dispose();
}
fdInstanceType = fdModelName + layout;
fldInstanceType = fldModelName + layout;
fdInstance = constructNetObject(fdInstanceType);
fldInstance = constructNetObject(fldInstanceType);
fdInputOptions = fdInstance.inputOptions;
fldInputOptions = fldInstance.inputOptions;
fdOutputs = {};
for (const outputInfo of Object.entries(fdInstance.outputsInfo)) {
fdOutputs[outputInfo[0]] =
new Float32Array(utils.sizeOfShape(outputInfo[1]));
}
fldOutputs = {'output': new Float32Array(utils.sizeOfShape([1, 136]))};
isFirstTimeLoad = false;
console.log(`- Model name: ${fdModelName}, Model layout: ${layout} -`);
// UI shows model loading progress
await ui.showProgressComponent('current', 'pending', 'pending');
console.log('- Loading weights... ');
const contextOptions = {'devicePreference': deviceType};
if (powerPreference) {
contextOptions['powerPreference'] = powerPreference;
}
if (numThreads) {
contextOptions['numThreads'] = numThreads;
}
start = performance.now();
const fdOutputOperand = await fdInstance.load(contextOptions);
const fldOutputOperand = await fldInstance.load(contextOptions);
loadTime = (performance.now() - start).toFixed(2);
console.log(` done in ${loadTime} ms.`);
// UI shows model building progress
await ui.showProgressComponent('done', 'current', 'pending');
console.log('- Building... ');
start = performance.now();
await fdInstance.build(fdOutputOperand);
await fldInstance.build(fldOutputOperand);
buildTime = (performance.now() - start).toFixed(2);
console.log(` done in ${buildTime} ms.`);
}
// UI shows inferencing progress
await ui.showProgressComponent('done', 'done', 'current');
if (inputType === 'image') {
const computeTimeArray = [];
let strokedRects;
let keyPoints;
let medianComputeTime;
console.log('- Computing... ');
// Do warm up
const fdResults = await fdInstance.compute(new Float32Array(
utils.sizeOfShape(fdInputOptions.inputDimensions)), fdOutputs);
const fldResults = await fldInstance.compute(new Float32Array(
utils.sizeOfShape(fldInputOptions.inputDimensions)), fldOutputs);
fdOutputs = fdResults.outputs;
fldOutputs = fldResults.outputs;
for (let i = 0; i < numRuns; i++) {
[computeTime, strokedRects, keyPoints] = await predict(imgElement);
console.log(` compute time ${i+1}: ${computeTime} ms`);
computeTimeArray.push(Number(computeTime));
}
if (numRuns > 1) {
medianComputeTime = utils.getMedianValue(computeTimeArray);
medianComputeTime = medianComputeTime.toFixed(2);
console.log(` median compute time: ${medianComputeTime} ms`);
}
console.log('Face Detection model outputs: ', fdOutputs);
console.log('Face Landmark model outputs: ', fldOutputs);
await ui.showProgressComponent('done', 'done', 'done');
$('#fps').hide();
ui.readyShowResultComponents();
await drawOutput(imgElement, strokedRects, keyPoints);
showPerfResult(medianComputeTime);
} else if (inputType === 'camera') {
stream = await utils.getMediaStream();
camElement.srcObject = stream;
camElement.onloadeddata = await renderCamStream();
await ui.showProgressComponent('done', 'done', 'done');
$('#fps').show();
ui.readyShowResultComponents();
} else {
throw Error(`Unknown inputType ${inputType}`);
}
} catch (error) {
console.log(error);
ui.addAlert(error.message);
}
ui.handleClick(disabledSelectors, false);
}