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

modularize human class and add model validation

pull/356/head
Vladimir Mandic 2021-09-12 18:37:06 -04:00
parent ba8ac1d8b8
commit fd0f85a8e9
12 changed files with 438 additions and 374 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CHANGELOG.md
View File

@ -11,6 +11,8 @@
### **HEAD -> main** 2021/09/12 mandic00@live.com
- add dynamic kernel op detection
- added human.env diagnostic class
- minor typos
- release candidate
- parametrize face config

2
README.md
View File

@ -47,7 +47,7 @@ Check out [**Live Demo**](https://vladmandic.github.io/human/demo/index.html) ap
- [**Code Repository**](https://github.com/vladmandic/human)
- [**NPM Package**](https://www.npmjs.com/package/@vladmandic/human)
- [**Issues Tracker**](https://github.com/vladmandic/human/issues)
- [**TypeDoc API Specification: Human**](https://vladmandic.github.io/human/typedoc/classes/Human.html)
- [**TypeDoc API Specification**](https://vladmandic.github.io/human/typedoc/classes/Human.html)
- [**Change Log**](https://github.com/vladmandic/human/blob/main/CHANGELOG.md)
- [**Current To-do List**](https://github.com/vladmandic/human/blob/main/TODO.md)

5
src/README.md
View File

@ -6,6 +6,5 @@ Source code of the `Human` library
- Compiled typings are present in `/types`
- Extracted API specification is present in `/typedoc`
For details how to build the `Human` library see Wiki
- [**Build Process**](https://github.com/vladmandic/human/wiki/Build-Process)
[**Build Process**](https://github.com/vladmandic/human/wiki/Build-Process)
[**TypeDoc API Specification**](https://vladmandic.github.io/human/typedoc/classes/Human.html)

338
src/human.ts
View File

@ -6,7 +6,6 @@ import { log, now, mergeDeep } from './helpers';
import { Config, defaults } from './config';
import { Result, FaceResult, HandResult, BodyResult, ObjectResult, GestureResult } from './result';
import * as tf from '../dist/tfjs.esm.js';
import * as backend from './tfjs/backend';
import * as models from './models';
import * as face from './face';
import * as facemesh from './blazeface/facemesh';
@ -24,9 +23,10 @@ import * as image from './image/image';
import * as draw from './draw/draw';
import * as persons from './persons';
import * as interpolate from './interpolate';
import * as sample from './sample';
import * as env from './env';
import * as backend from './tfjs/backend';
import * as app from '../package.json';
import * as warmups from './warmup';
import { Tensor, GraphModel } from './tfjs/types';
// export types
@ -86,8 +86,6 @@ export class Human {
* - Progresses through: 'config', 'check', 'backend', 'load', 'run:<model>', 'idle'
*/
state: string;
/** process input and return tensor and canvas */
image: typeof image.process;
/** currenty processed image tensor and canvas */
process: { tensor: Tensor | null, canvas: OffscreenCanvas | HTMLCanvasElement | null };
/** @internal: Instance of TensorFlow/JS used by Human
@ -151,9 +149,7 @@ export class Human {
#numTensors: number;
#analyzeMemoryLeaks: boolean;
#checkSanity: boolean;
#firstRun: boolean;
#lastInputSum: number;
#lastCacheDiff: number;
initial: boolean;
// definition end
@ -176,18 +172,17 @@ export class Human {
this.#numTensors = 0;
this.#analyzeMemoryLeaks = false;
this.#checkSanity = false;
this.#firstRun = true;
this.#lastCacheDiff = 0;
this.initial = true;
this.performance = { backend: 0, load: 0, image: 0, frames: 0, cached: 0, changed: 0, total: 0, draw: 0 };
this.events = new EventTarget();
// object that contains all initialized models
this.models = {
face: null,
face: null, // array of models
posenet: null,
blazepose: null,
efficientpose: null,
movenet: null,
handpose: null,
handpose: null, // array of models
age: null,
gender: null,
emotion: null,
@ -200,14 +195,12 @@ export class Human {
this.result = { face: [], body: [], hand: [], gesture: [], object: [], performance: {}, timestamp: 0, persons: [] };
// export access to image this.processing
// @ts-ignore eslint-typescript cannot correctly infer type in anonymous function
this.image = (input: Input) => image.process(input, this.config);
this.process = { tensor: null, canvas: null };
// export raw access to underlying models
this.faceTriangulation = facemesh.triangulation;
this.faceUVMap = facemesh.uvmap;
// include platform info
this.#lastInputSum = 1;
this.#emit('create');
this.emit('create');
}
// helper function: measure tensor leak
@ -235,6 +228,13 @@ export class Human {
return null;
}
/** Process input as return canvas and tensor
*
* @param input: {@link Input}
* @returns { tensor, canvas }
*/
image = (input: Input) => image.process(input, this.config);
/** Simmilarity method calculates simmilarity between two provided face descriptors (face embeddings)
* - Calculation is based on normalized Minkowski distance between
*
@ -290,12 +290,11 @@ export class Human {
const count = Object.values(this.models).filter((model) => model).length;
if (userConfig) this.config = mergeDeep(this.config, userConfig) as Config;
if (this.#firstRun) { // print version info on first run and check for correct backend setup
if (this.initial) { // print version info on first run and check for correct backend setup
if (this.config.debug) log(`version: ${this.version}`);
if (this.config.debug) log(`tfjs version: ${this.tf.version_core}`);
// if (this.config.debug) log('environment:', this.env);
await this.#checkBackend();
await backend.check(this);
await tf.ready();
if (this.env.browser) {
if (this.config.debug) log('configuration:', this.config);
if (this.config.debug) log('tf flags:', this.tf.ENV.flags);
@ -304,123 +303,22 @@ export class Human {
await models.load(this); // actually loads models
if (this.#firstRun) { // print memory stats on first run
if (this.config.debug) log('tf engine state:', this.tf.engine().state.numBytes, 'bytes', this.tf.engine().state.numTensors, 'tensors');
this.#firstRun = false;
}
if (this.initial && this.config.debug) log('tf engine state:', this.tf.engine().state.numBytes, 'bytes', this.tf.engine().state.numTensors, 'tensors'); // print memory stats on first run
this.initial = false;
const loaded = Object.values(this.models).filter((model) => model).length;
if (loaded !== count) this.#emit('load');
if (loaded !== count) { // number of loaded models changed
await models.validate(this); // validate kernel ops used by model against current backend
this.emit('load');
}
const current = Math.trunc(now() - timeStamp);
if (current > (this.performance.load as number || 0)) this.performance.load = current;
}
// emit event
/** @hidden */
#emit = (event: string) => this.events?.dispatchEvent(new Event(event));
// check if backend needs initialization if it changed
/** @hidden */
#checkBackend = async () => {
if (this.#firstRun || (this.config.backend && (this.config.backend.length > 0) && (this.tf.getBackend() !== this.config.backend))) {
const timeStamp = now();
this.state = 'backend';
/* force backend reload
if (this.config.backend in tf.engine().registry) {
const backendFactory = tf.findBackendFactory(this.config.backend);
tf.removeBackend(this.config.backend);
tf.registerBackend(this.config.backend, backendFactory);
} else {
log('Backend not registred:', this.config.backend);
}
*/
if (this.config.backend && this.config.backend.length > 0) {
// detect web worker
// @ts-ignore ignore missing type for WorkerGlobalScope as that is the point
if (typeof window === 'undefined' && typeof WorkerGlobalScope !== 'undefined' && this.config.debug) {
log('running inside web worker');
}
// force browser vs node backend
if (this.env.browser && this.config.backend === 'tensorflow') {
log('override: backend set to tensorflow while running in browser');
this.config.backend = 'humangl';
}
if (this.env.node && (this.config.backend === 'webgl' || this.config.backend === 'humangl')) {
log(`override: backend set to ${this.config.backend} while running in nodejs`);
this.config.backend = 'tensorflow';
}
// handle webgpu
if (this.env.browser && this.config.backend === 'webgpu') {
if (typeof navigator === 'undefined' || typeof navigator['gpu'] === 'undefined') {
log('override: backend set to webgpu but browser does not support webgpu');
this.config.backend = 'humangl';
} else {
const adapter = await navigator['gpu'].requestAdapter();
if (this.config.debug) log('enumerated webgpu adapter:', adapter);
}
}
// check available backends
if (this.config.backend === 'humangl') backend.register();
const available = Object.keys(this.tf.engine().registryFactory);
if (this.config.debug) log('available backends:', available);
if (!available.includes(this.config.backend)) {
log(`error: backend ${this.config.backend} not found in registry`);
this.config.backend = this.env.node ? 'tensorflow' : 'humangl';
log(`override: setting backend ${this.config.backend}`);
}
if (this.config.debug) log('setting backend:', this.config.backend);
// handle wasm
if (this.config.backend === 'wasm') {
if (this.config.debug) log('wasm path:', this.config.wasmPath);
if (typeof this.tf?.setWasmPaths !== 'undefined') this.tf.setWasmPaths(this.config.wasmPath);
else throw new Error('Human: WASM backend is not loaded');
const simd = await this.tf.env().getAsync('WASM_HAS_SIMD_SUPPORT');
const mt = await this.tf.env().getAsync('WASM_HAS_MULTITHREAD_SUPPORT');
if (this.config.debug) log(`wasm execution: ${simd ? 'SIMD' : 'no SIMD'} ${mt ? 'multithreaded' : 'singlethreaded'}`);
if (this.config.debug && !simd) log('warning: wasm simd support is not enabled');
}
// handle humangl
try {
await this.tf.setBackend(this.config.backend);
} catch (err) {
log('error: cannot set backend:', this.config.backend, err);
}
}
// handle webgl & humangl
if (this.tf.getBackend() === 'humangl') {
this.tf.ENV.set('CHECK_COMPUTATION_FOR_ERRORS', false);
this.tf.ENV.set('WEBGL_CPU_FORWARD', true);
this.tf.ENV.set('WEBGL_PACK_DEPTHWISECONV', false);
this.tf.ENV.set('WEBGL_USE_SHAPES_UNIFORMS', true);
// if (!this.config.object.enabled) this.tf.ENV.set('WEBGL_FORCE_F16_TEXTURES', true); // safe to use 16bit precision
if (typeof this.config['deallocate'] !== 'undefined' && this.config['deallocate']) { // hidden param
log('changing webgl: WEBGL_DELETE_TEXTURE_THRESHOLD:', true);
this.tf.ENV.set('WEBGL_DELETE_TEXTURE_THRESHOLD', 0);
}
// @ts-ignore getGPGPUContext only exists on WebGL backend
const gl = await this.tf.backend().getGPGPUContext().gl;
if (this.config.debug) log(`gl version:${gl.getParameter(gl.VERSION)} renderer:${gl.getParameter(gl.RENDERER)}`);
}
// wait for ready
this.tf.enableProdMode();
await this.tf.ready();
this.performance.backend = Math.trunc(now() - timeStamp);
this.config.backend = this.tf.getBackend();
env.get(); // update env on backend init
this.env = env.env;
}
}
emit = (event: string) => this.events?.dispatchEvent(new Event(event));
/**
* Runs interpolation using last known result and returns smoothened result
@ -431,43 +329,20 @@ export class Human {
*/
next = (result?: Result) => interpolate.calc(result || this.result) as Result;
// check if input changed sufficiently to trigger new detections
/** @hidden */
#skipFrame = async (input: Tensor) => {
if (this.config.cacheSensitivity === 0) return false;
const resizeFact = 32;
if (!input.shape[1] || !input.shape[2]) return false;
const reduced: Tensor = tf.image.resizeBilinear(input, [Math.trunc(input.shape[1] / resizeFact), Math.trunc(input.shape[2] / resizeFact)]);
// use tensor sum
/*
const sumT = this.tf.sum(reduced);
const sum = await sumT.data()[0] as number;
sumT.dispose();
/** Warmup method pre-initializes all configured models for faster inference
* - can take significant time on startup
* - only used for `webgl` and `humangl` backends
* @param userConfig?: {@link Config}
*/
// use js loop sum, faster than uploading tensor to gpu calculating and downloading back
const reducedData = await reduced.data(); // raw image rgb array
let sum = 0;
for (let i = 0; i < reducedData.length / 3; i++) sum += reducedData[3 * i + 2]; // look only at green value of each pixel
reduced.dispose();
const diff = 100 * (Math.max(sum, this.#lastInputSum) / Math.min(sum, this.#lastInputSum) - 1);
this.#lastInputSum = sum;
// if previous frame was skipped, skip this frame if changed more than cacheSensitivity
// if previous frame was not skipped, then look for cacheSensitivity or difference larger than one in previous frame to avoid resetting cache in subsequent frames unnecessarily
const skipFrame = diff < Math.max(this.config.cacheSensitivity, this.#lastCacheDiff);
// if difference is above 10x threshold, don't use last value to force reset cache for significant change of scenes or images
this.#lastCacheDiff = diff > 10 * this.config.cacheSensitivity ? 0 : diff;
// console.log('skipFrame', skipFrame, this.config.cacheSensitivity, diff);
return skipFrame;
}
warmup = (userConfig?: Partial<Config>) => warmups.warmup(this, userConfig) as Promise<Result | { error }>
/** Main detection method
* - Analyze configuration: {@link Config}
* - Pre-this.process input: {@link Input}
* - Run inference for all configured models
* - this.process and return result: {@link Result}
* - Process and return result: {@link Result}
*
* @param input: Input
* @param input: {@link Input}
* @param userConfig?: {@link Config}
* @returns result: {@link Result}
*/
@ -491,19 +366,20 @@ export class Human {
const timeStart = now();
// configure backend
await this.#checkBackend();
// configure backend if needed
await backend.check(this);
// load models if enabled
await this.load();
timeStamp = now();
this.process = image.process(input, this.config);
const inputTensor = this.process.tensor;
this.performance.image = Math.trunc(now() - timeStamp);
this.analyze('Get Image:');
// run segmentation prethis.processing
if (this.config.segmentation.enabled && this.process && this.process.tensor) {
if (this.config.segmentation.enabled && this.process && inputTensor) {
this.analyze('Start Segmentation:');
this.state = 'run:segmentation';
timeStamp = now();
@ -512,21 +388,21 @@ export class Human {
if (elapsedTime > 0) this.performance.segmentation = elapsedTime;
if (this.process.canvas) {
// replace input
tf.dispose(this.process.tensor);
tf.dispose(inputTensor);
this.process = image.process(this.process.canvas, this.config);
}
this.analyze('End Segmentation:');
}
if (!this.process || !this.process.tensor) {
if (!this.process || !inputTensor) {
log('could not convert input to tensor');
resolve({ error: 'could not convert input to tensor' });
return;
}
this.#emit('image');
this.emit('image');
timeStamp = now();
this.config.skipFrame = await this.#skipFrame(this.process.tensor);
this.config.skipFrame = await image.skip(this, inputTensor);
if (!this.performance.frames) this.performance.frames = 0;
if (!this.performance.cached) this.performance.cached = 0;
(this.performance.frames as number)++;
@ -543,12 +419,12 @@ export class Human {
// run face detection followed by all models that rely on face bounding box: face mesh, age, gender, emotion
if (this.config.async) {
faceRes = this.config.face.enabled ? face.detectFace(this, this.process.tensor) : [];
faceRes = this.config.face.enabled ? face.detectFace(this, inputTensor) : [];
if (this.performance.face) delete this.performance.face;
} else {
this.state = 'run:face';
timeStamp = now();
faceRes = this.config.face.enabled ? await face.detectFace(this, this.process.tensor) : [];
faceRes = this.config.face.enabled ? await face.detectFace(this, inputTensor) : [];
elapsedTime = Math.trunc(now() - timeStamp);
if (elapsedTime > 0) this.performance.face = elapsedTime;
}
@ -556,18 +432,18 @@ export class Human {
// run body: can be posenet, blazepose, efficientpose, movenet
this.analyze('Start Body:');
if (this.config.async) {
if (this.config.body.modelPath?.includes('posenet')) bodyRes = this.config.body.enabled ? posenet.predict(this.process.tensor, this.config) : [];
else if (this.config.body.modelPath?.includes('blazepose')) bodyRes = this.config.body.enabled ? blazepose.predict(this.process.tensor, this.config) : [];
else if (this.config.body.modelPath?.includes('efficientpose')) bodyRes = this.config.body.enabled ? efficientpose.predict(this.process.tensor, this.config) : [];
else if (this.config.body.modelPath?.includes('movenet')) bodyRes = this.config.body.enabled ? movenet.predict(this.process.tensor, this.config) : [];
if (this.config.body.modelPath?.includes('posenet')) bodyRes = this.config.body.enabled ? posenet.predict(inputTensor, this.config) : [];
else if (this.config.body.modelPath?.includes('blazepose')) bodyRes = this.config.body.enabled ? blazepose.predict(inputTensor, this.config) : [];
else if (this.config.body.modelPath?.includes('efficientpose')) bodyRes = this.config.body.enabled ? efficientpose.predict(inputTensor, this.config) : [];
else if (this.config.body.modelPath?.includes('movenet')) bodyRes = this.config.body.enabled ? movenet.predict(inputTensor, this.config) : [];
if (this.performance.body) delete this.performance.body;
} else {
this.state = 'run:body';
timeStamp = now();
if (this.config.body.modelPath?.includes('posenet')) bodyRes = this.config.body.enabled ? await posenet.predict(this.process.tensor, this.config) : [];
else if (this.config.body.modelPath?.includes('blazepose')) bodyRes = this.config.body.enabled ? await blazepose.predict(this.process.tensor, this.config) : [];
else if (this.config.body.modelPath?.includes('efficientpose')) bodyRes = this.config.body.enabled ? await efficientpose.predict(this.process.tensor, this.config) : [];
else if (this.config.body.modelPath?.includes('movenet')) bodyRes = this.config.body.enabled ? await movenet.predict(this.process.tensor, this.config) : [];
if (this.config.body.modelPath?.includes('posenet')) bodyRes = this.config.body.enabled ? await posenet.predict(inputTensor, this.config) : [];
else if (this.config.body.modelPath?.includes('blazepose')) bodyRes = this.config.body.enabled ? await blazepose.predict(inputTensor, this.config) : [];
else if (this.config.body.modelPath?.includes('efficientpose')) bodyRes = this.config.body.enabled ? await efficientpose.predict(inputTensor, this.config) : [];
else if (this.config.body.modelPath?.includes('movenet')) bodyRes = this.config.body.enabled ? await movenet.predict(inputTensor, this.config) : [];
elapsedTime = Math.trunc(now() - timeStamp);
if (elapsedTime > 0) this.performance.body = elapsedTime;
}
@ -576,12 +452,12 @@ export class Human {
// run handpose
this.analyze('Start Hand:');
if (this.config.async) {
handRes = this.config.hand.enabled ? handpose.predict(this.process.tensor, this.config) : [];
handRes = this.config.hand.enabled ? handpose.predict(inputTensor, this.config) : [];
if (this.performance.hand) delete this.performance.hand;
} else {
this.state = 'run:hand';
timeStamp = now();
handRes = this.config.hand.enabled ? await handpose.predict(this.process.tensor, this.config) : [];
handRes = this.config.hand.enabled ? await handpose.predict(inputTensor, this.config) : [];
elapsedTime = Math.trunc(now() - timeStamp);
if (elapsedTime > 0) this.performance.hand = elapsedTime;
}
@ -590,14 +466,14 @@ export class Human {
// run nanodet
this.analyze('Start Object:');
if (this.config.async) {
if (this.config.object.modelPath?.includes('nanodet')) objectRes = this.config.object.enabled ? nanodet.predict(this.process.tensor, this.config) : [];
else if (this.config.object.modelPath?.includes('centernet')) objectRes = this.config.object.enabled ? centernet.predict(this.process.tensor, this.config) : [];
if (this.config.object.modelPath?.includes('nanodet')) objectRes = this.config.object.enabled ? nanodet.predict(inputTensor, this.config) : [];
else if (this.config.object.modelPath?.includes('centernet')) objectRes = this.config.object.enabled ? centernet.predict(inputTensor, this.config) : [];
if (this.performance.object) delete this.performance.object;
} else {
this.state = 'run:object';
timeStamp = now();
if (this.config.object.modelPath?.includes('nanodet')) objectRes = this.config.object.enabled ? await nanodet.predict(this.process.tensor, this.config) : [];
else if (this.config.object.modelPath?.includes('centernet')) objectRes = this.config.object.enabled ? await centernet.predict(this.process.tensor, this.config) : [];
if (this.config.object.modelPath?.includes('nanodet')) objectRes = this.config.object.enabled ? await nanodet.predict(inputTensor, this.config) : [];
else if (this.config.object.modelPath?.includes('centernet')) objectRes = this.config.object.enabled ? await centernet.predict(inputTensor, this.config) : [];
elapsedTime = Math.trunc(now() - timeStamp);
if (elapsedTime > 0) this.performance.object = elapsedTime;
}
@ -631,111 +507,13 @@ export class Human {
};
// finally dispose input tensor
tf.dispose(this.process.tensor);
tf.dispose(inputTensor);
// log('Result:', result);
this.#emit('detect');
this.emit('detect');
resolve(this.result);
});
}
/** @hidden */
#warmupBitmap = async () => {
const b64toBlob = (base64: string, type = 'application/octet-stream') => fetch(`data:${type};base64,${base64}`).then((res) => res.blob());
let blob;
let res;
switch (this.config.warmup) {
case 'face': blob = await b64toBlob(sample.face); break;
case 'full': blob = await b64toBlob(sample.body); break;
default: blob = null;
}
if (blob) {
const bitmap = await createImageBitmap(blob);
res = await this.detect(bitmap, this.config);
bitmap.close();
}
return res;
}
/** @hidden */
#warmupCanvas = async () => new Promise((resolve) => {
let src;
let size = 0;
switch (this.config.warmup) {
case 'face':
size = 256;
src = 'data:image/jpeg;base64,' + sample.face;
break;
case 'full':
case 'body':
size = 1200;
src = 'data:image/jpeg;base64,' + sample.body;
break;
default:
src = null;
}
// src = encodeURI('../assets/human-sample-upper.jpg');
const img = new Image();
img.onload = async () => {
const canvas = (typeof OffscreenCanvas !== 'undefined') ? new OffscreenCanvas(size, size) : document.createElement('canvas');
canvas.width = img.naturalWidth;
canvas.height = img.naturalHeight;
const ctx = canvas.getContext('2d');
ctx?.drawImage(img, 0, 0);
// const data = ctx?.getImageData(0, 0, canvas.height, canvas.width);
const res = await this.detect(canvas, this.config);
resolve(res);
};
if (src) img.src = src;
else resolve(null);
});
/** @hidden */
#warmupNode = async () => {
const atob = (str: string) => Buffer.from(str, 'base64');
let img;
if (this.config.warmup === 'face') img = atob(sample.face);
if (this.config.warmup === 'body' || this.config.warmup === 'full') img = atob(sample.body);
if (!img) return null;
let res;
if (typeof tf['node'] !== 'undefined') {
const data = tf['node'].decodeJpeg(img);
const expanded = data.expandDims(0);
this.tf.dispose(data);
// log('Input:', expanded);
res = await this.detect(expanded, this.config);
this.tf.dispose(expanded);
} else {
if (this.config.debug) log('Warmup tfjs-node not loaded');
/*
const input = await canvasJS.loadImage(img);
const canvas = canvasJS.createCanvas(input.width, input.height);
const ctx = canvas.getContext('2d');
ctx.drawImage(img, 0, 0, input.width, input.height);
res = await this.detect(input, this.config);
*/
}
return res;
}
/** Warmup method pre-initializes all configured models for faster inference
* - can take significant time on startup
* - only used for `webgl` and `humangl` backends
* @param userConfig?: Config
*/
async warmup(userConfig?: Partial<Config>): Promise<Result | { error }> {
const t0 = now();
if (userConfig) this.config = mergeDeep(this.config, userConfig) as Config;
if (!this.config.warmup || this.config.warmup === 'none') return { error: 'null' };
let res;
if (typeof createImageBitmap === 'function') res = await this.#warmupBitmap();
else if (typeof Image !== 'undefined') res = await this.#warmupCanvas();
else res = await this.#warmupNode();
const t1 = now();
if (this.config.debug) log('Warmup', this.config.warmup, Math.round(t1 - t0), 'ms', res);
this.#emit('warmup');
return res;
}
}
/**

30
src/image/image.ts
View File

@ -177,3 +177,33 @@ export function process(input: Input, config: Config): { tensor: Tensor | null,
const canvas = config.filter.return ? outCanvas : null;
return { tensor, canvas };
}
let lastInputSum = 0;
let lastCacheDiff = 1;
export async function skip(instance, input: Tensor) {
if (instance.config.cacheSensitivity === 0) return false;
const resizeFact = 32;
if (!input.shape[1] || !input.shape[2]) return false;
const reduced: Tensor = tf.image.resizeBilinear(input, [Math.trunc(input.shape[1] / resizeFact), Math.trunc(input.shape[2] / resizeFact)]);
// use tensor sum
/*
const sumT = this.tf.sum(reduced);
const sum = await sumT.data()[0] as number;
sumT.dispose();
*/
// use js loop sum, faster than uploading tensor to gpu calculating and downloading back
const reducedData = await reduced.data(); // raw image rgb array
let sum = 0;
for (let i = 0; i < reducedData.length / 3; i++) sum += reducedData[3 * i + 2]; // look only at green value of each pixel
reduced.dispose();
const diff = 100 * (Math.max(sum, lastInputSum) / Math.min(sum, lastInputSum) - 1);
lastInputSum = sum;
// if previous frame was skipped, skip this frame if changed more than cacheSensitivity
// if previous frame was not skipped, then look for cacheSensitivity or difference larger than one in previous frame to avoid resetting cache in subsequent frames unnecessarily
const skipFrame = diff < Math.max(instance.config.cacheSensitivity, lastCacheDiff);
// if difference is above 10x threshold, don't use last value to force reset cache for significant change of scenes or images
lastCacheDiff = diff > 10 * instance.config.cacheSensitivity ? 0 : diff;
// console.log('skipFrame', skipFrame, this.config.cacheSensitivity, diff);
return skipFrame;
}

38
src/models.ts
View File

@ -1,3 +1,5 @@
import { log } from './helpers';
import { GraphModel } from './tfjs/types';
import * as facemesh from './blazeface/facemesh';
import * as faceres from './faceres/faceres';
import * as emotion from './emotion/emotion';
@ -59,3 +61,39 @@ export async function load(instance) {
// if (instance.config.face.enabled && instance.config.face.agegenderrace.enabled && !instance.models.agegenderrace) instance.models.agegenderrace = await agegenderrace.load(instance.config);
}
}
export async function validate(instance) {
interface Op { name: string, category: string, op: string }
const simpleOps = ['const', 'placeholder', 'noop', 'pad', 'squeeze', 'add', 'sub', 'mul', 'div'];
for (const defined of Object.keys(instance.models)) {
if (instance.models[defined]) { // check if model is loaded
let models: GraphModel[] = [];
if (Array.isArray(instance.models[defined])) models = instance.models[defined].map((model) => (model.executor ? model : model.model));
else models = [instance.models[defined]];
for (const model of models) {
const ops: string[] = [];
// @ts-ignore // executor is a private method
const executor = model?.executor;
if (executor) {
for (const kernel of Object.values(executor.graph.nodes)) {
const op = (kernel as Op).op.toLowerCase();
if (!ops.includes(op)) ops.push(op);
}
}
const missing: string[] = [];
for (const op of ops) {
if (!simpleOps.includes(op) // exclude simple ops
&& !instance.env.kernels.includes(op) // check actual kernel ops
&& !instance.env.kernels.includes(op.replace('_', '')) // check variation without _
&& !instance.env.kernels.includes(op.replace('native', '')) // check standard variation
&& !instance.env.kernels.includes(op.replace('v2', ''))) { // check non-versioned variation
missing.push(op);
}
}
if (!executor && instance.config.debug) log('model executor not found:', defined);
if (missing.length > 0 && instance.config.debug) log('model validation:', defined, missing);
}
}
}
// log.data('ops used by model:', ops);
}

181
src/tfjs/backend.ts
View File

@ -1,92 +1,107 @@
/**
* Custom TFJS backend for Human based on WebGL
* Not used by default
*/
import { log } from '../helpers';
import { log, now } from '../helpers';
import * as humangl from './humangl';
import * as env from '../env';
import * as tf from '../../dist/tfjs.esm.js';
export const config = {
name: 'humangl',
priority: 99,
canvas: <null | OffscreenCanvas | HTMLCanvasElement>null,
gl: <null | WebGL2RenderingContext>null,
width: 1024,
height: 1024,
extensions: <string[]> [],
webGLattr: { // https://www.khronos.org/registry/webgl/specs/latest/1.0/#5.2
alpha: false,
antialias: false,
premultipliedAlpha: false,
preserveDrawingBuffer: false,
depth: false,
stencil: false,
failIfMajorPerformanceCaveat: false,
desynchronized: true,
},
};
function extensions(): void {
/*
https://www.khronos.org/registry/webgl/extensions/
https://webglreport.com/?v=2
export async function check(instance) {
if (instance.initial || (instance.config.backend && (instance.config.backend.length > 0) && (tf.getBackend() !== instance.config.backend))) {
const timeStamp = now();
instance.state = 'backend';
/* force backend reload
if (instance.config.backend in tf.engine().registry) {
const backendFactory = tf.findBackendFactory(instance.config.backend);
tf.removeBackend(instance.config.backend);
tf.registerBackend(instance.config.backend, backendFactory);
} else {
log('Backend not registred:', instance.config.backend);
}
*/
const gl = config.gl;
if (!gl) return;
config.extensions = gl.getSupportedExtensions() as string[];
// gl.getExtension('KHR_parallel_shader_compile');
if (instance.config.backend && instance.config.backend.length > 0) {
// detect web worker
// @ts-ignore ignore missing type for WorkerGlobalScope as that is the point
if (typeof window === 'undefined' && typeof WorkerGlobalScope !== 'undefined' && instance.config.debug) {
log('running inside web worker');
}
/**
* Registers custom WebGL2 backend to be used by Human library
*
* @returns void
*/
export function register(): void {
if (!tf.findBackend(config.name)) {
// log('backend registration:', config.name);
// force browser vs node backend
if (env.env.browser && instance.config.backend === 'tensorflow') {
log('override: backend set to tensorflow while running in browser');
instance.config.backend = 'humangl';
}
if (env.env.node && (instance.config.backend === 'webgl' || instance.config.backend === 'humangl')) {
log(`override: backend set to ${instance.config.backend} while running in nodejs`);
instance.config.backend = 'tensorflow';
}
// handle webgpu
if (env.env.browser && instance.config.backend === 'webgpu') {
if (typeof navigator === 'undefined' || typeof navigator['gpu'] === 'undefined') {
log('override: backend set to webgpu but browser does not support webgpu');
instance.config.backend = 'humangl';
} else {
const adapter = await navigator['gpu'].requestAdapter();
if (instance.config.debug) log('enumerated webgpu adapter:', adapter);
}
}
// check available backends
if (instance.config.backend === 'humangl') humangl.register();
const available = Object.keys(tf.engine().registryFactory);
if (instance.config.debug) log('available backends:', available);
if (!available.includes(instance.config.backend)) {
log(`error: backend ${instance.config.backend} not found in registry`);
instance.config.backend = env.env.node ? 'tensorflow' : 'humangl';
log(`override: setting backend ${instance.config.backend}`);
}
if (instance.config.debug) log('setting backend:', instance.config.backend);
// handle wasm
if (instance.config.backend === 'wasm') {
if (instance.config.debug) log('wasm path:', instance.config.wasmPath);
if (typeof tf?.setWasmPaths !== 'undefined') await tf.setWasmPaths(instance.config.wasmPath);
else throw new Error('Human: WASM backend is not loaded');
const simd = await tf.env().getAsync('WASM_HAS_SIMD_SUPPORT');
const mt = await tf.env().getAsync('WASM_HAS_MULTITHREAD_SUPPORT');
if (instance.config.debug) log(`wasm execution: ${simd ? 'SIMD' : 'no SIMD'} ${mt ? 'multithreaded' : 'singlethreaded'}`);
if (instance.config.debug && !simd) log('warning: wasm simd support is not enabled');
}
await tf.setBackend(instance.config.backend);
try {
config.canvas = (typeof OffscreenCanvas !== 'undefined') ? new OffscreenCanvas(config.width, config.height) : document.createElement('canvas');
await tf.setBackend(instance.config.backend);
await tf.ready();
} catch (err) {
log('error: cannot create canvas:', err);
return;
}
try {
config.gl = config.canvas.getContext('webgl2', config.webGLattr) as WebGL2RenderingContext;
} catch (err) {
log('error: cannot get WebGL2 context:', err);
return;
}
try {
tf.setWebGLContext(2, config.gl);
} catch (err) {
log('error: cannot set WebGL2 context:', err);
return;
}
try {
const ctx = new tf.GPGPUContext(config.gl);
tf.registerBackend(config.name, () => new tf.MathBackendWebGL(ctx), config.priority);
} catch (err) {
log('error: cannot register WebGL backend:', err);
return;
}
try {
const kernels = tf.getKernelsForBackend('webgl');
kernels.forEach((kernelConfig) => {
const newKernelConfig = { ...kernelConfig, backendName: config.name };
tf.registerKernel(newKernelConfig);
});
} catch (err) {
log('error: cannot update WebGL backend registration:', err);
return;
}
try {
tf.ENV.set('WEBGL_VERSION', 2);
} catch (err) {
log('error: cannot set WebGL backend flags:', err);
return;
}
extensions();
log('backend registered:', config.name);
log('error: cannot set backend:', instance.config.backend, err);
}
}
// handle webgl & humangl
if (tf.getBackend() === 'humangl') {
tf.ENV.set('CHECK_COMPUTATION_FOR_ERRORS', false);
tf.ENV.set('WEBGL_CPU_FORWARD', true);
tf.ENV.set('WEBGL_PACK_DEPTHWISECONV', false);
tf.ENV.set('WEBGL_USE_SHAPES_UNIFORMS', true);
// if (!instance.config.object.enabled) tf.ENV.set('WEBGL_FORCE_F16_TEXTURES', true); // safe to use 16bit precision
if (typeof instance.config['deallocate'] !== 'undefined' && instance.config['deallocate']) { // hidden param
log('changing webgl: WEBGL_DELETE_TEXTURE_THRESHOLD:', true);
tf.ENV.set('WEBGL_DELETE_TEXTURE_THRESHOLD', 0);
}
// @ts-ignore getGPGPUContext only exists on WebGL backend
const gl = await tf.backend().getGPGPUContext().gl;
if (instance.config.debug) log(`gl version:${gl.getParameter(gl.VERSION)} renderer:${gl.getParameter(gl.RENDERER)}`);
}
// wait for ready
tf.enableProdMode();
await tf.ready();
instance.performance.backend = Math.trunc(now() - timeStamp);
instance.config.backend = tf.getBackend();
env.get(); // update env on backend init
instance.env = env.env;
}
}

92
src/tfjs/humangl.ts Normal file
View File

@ -0,0 +1,92 @@
/**
* Custom TFJS backend for Human based on WebGL
* Not used by default
*/
import { log } from '../helpers';
import * as tf from '../../dist/tfjs.esm.js';
export const config = {
name: 'humangl',
priority: 99,
canvas: <null | OffscreenCanvas | HTMLCanvasElement>null,
gl: <null | WebGL2RenderingContext>null,
width: 1024,
height: 1024,
extensions: <string[]> [],
webGLattr: { // https://www.khronos.org/registry/webgl/specs/latest/1.0/#5.2
alpha: false,
antialias: false,
premultipliedAlpha: false,
preserveDrawingBuffer: false,
depth: false,
stencil: false,
failIfMajorPerformanceCaveat: false,
desynchronized: true,
},
};
function extensions(): void {
/*
https://www.khronos.org/registry/webgl/extensions/
https://webglreport.com/?v=2
*/
const gl = config.gl;
if (!gl) return;
config.extensions = gl.getSupportedExtensions() as string[];
// gl.getExtension('KHR_parallel_shader_compile');
}
/**
* Registers custom WebGL2 backend to be used by Human library
*
* @returns void
*/
export function register(): void {
if (!tf.findBackend(config.name)) {
// log('backend registration:', config.name);
try {
config.canvas = (typeof OffscreenCanvas !== 'undefined') ? new OffscreenCanvas(config.width, config.height) : document.createElement('canvas');
} catch (err) {
log('error: cannot create canvas:', err);
return;
}
try {
config.gl = config.canvas.getContext('webgl2', config.webGLattr) as WebGL2RenderingContext;
} catch (err) {
log('error: cannot get WebGL2 context:', err);
return;
}
try {
tf.setWebGLContext(2, config.gl);
} catch (err) {
log('error: cannot set WebGL2 context:', err);
return;
}
try {
const ctx = new tf.GPGPUContext(config.gl);
tf.registerBackend(config.name, () => new tf.MathBackendWebGL(ctx), config.priority);
} catch (err) {
log('error: cannot register WebGL backend:', err);
return;
}
try {
const kernels = tf.getKernelsForBackend('webgl');
kernels.forEach((kernelConfig) => {
const newKernelConfig = { ...kernelConfig, backendName: config.name };
tf.registerKernel(newKernelConfig);
});
} catch (err) {
log('error: cannot update WebGL backend registration:', err);
return;
}
try {
tf.ENV.set('WEBGL_VERSION', 2);
} catch (err) {
log('error: cannot set WebGL backend flags:', err);
return;
}
extensions();
log('backend registered:', config.name);
}
}

102
src/warmup.ts Normal file
View File

@ -0,0 +1,102 @@
import { log, now, mergeDeep } from './helpers';
import * as sample from './sample';
import * as tf from '../dist/tfjs.esm.js';
import { Config } from './config';
import { Result } from './result';
async function warmupBitmap(instance) {
const b64toBlob = (base64: string, type = 'application/octet-stream') => fetch(`data:${type};base64,${base64}`).then((res) => res.blob());
let blob;
let res;
switch (instance.config.warmup) {
case 'face': blob = await b64toBlob(sample.face); break;
case 'full': blob = await b64toBlob(sample.body); break;
default: blob = null;
}
if (blob) {
const bitmap = await createImageBitmap(blob);
res = await instance.detect(bitmap, instance.config);
bitmap.close();
}
return res;
}
async function warmupCanvas(instance) {
return new Promise((resolve) => {
let src;
let size = 0;
switch (instance.config.warmup) {
case 'face':
size = 256;
src = 'data:image/jpeg;base64,' + sample.face;
break;
case 'full':
case 'body':
size = 1200;
src = 'data:image/jpeg;base64,' + sample.body;
break;
default:
src = null;
}
// src = encodeURI('../assets/human-sample-upper.jpg');
const img = new Image();
img.onload = async () => {
const canvas = (typeof OffscreenCanvas !== 'undefined') ? new OffscreenCanvas(size, size) : document.createElement('canvas');
canvas.width = img.naturalWidth;
canvas.height = img.naturalHeight;
const ctx = canvas.getContext('2d');
ctx?.drawImage(img, 0, 0);
// const data = ctx?.getImageData(0, 0, canvas.height, canvas.width);
const res = await instance.detect(canvas, instance.config);
resolve(res);
};
if (src) img.src = src;
else resolve(null);
});
}
async function warmupNode(instance) {
const atob = (str: string) => Buffer.from(str, 'base64');
let img;
if (instance.config.warmup === 'face') img = atob(sample.face);
if (instance.config.warmup === 'body' || instance.config.warmup === 'full') img = atob(sample.body);
if (!img) return null;
let res;
if (typeof tf['node'] !== 'undefined') {
const data = tf['node'].decodeJpeg(img);
const expanded = data.expandDims(0);
instance.tf.dispose(data);
// log('Input:', expanded);
res = await instance.detect(expanded, instance.config);
instance.tf.dispose(expanded);
} else {
if (instance.config.debug) log('Warmup tfjs-node not loaded');
/*
const input = await canvasJS.loadImage(img);
const canvas = canvasJS.createCanvas(input.width, input.height);
const ctx = canvas.getContext('2d');
ctx.drawImage(img, 0, 0, input.width, input.height);
res = await instance.detect(input, instance.config);
*/
}
return res;
}
/** Warmup method pre-initializes all configured models for faster inference
* - can take significant time on startup
* - only used for `webgl` and `humangl` backends
* @param userConfig?: Config
*/
export async function warmup(instance, userConfig?: Partial<Config>): Promise<Result | { error }> {
const t0 = now();
if (userConfig) instance.config = mergeDeep(instance.config, userConfig) as Config;
if (!instance.config.warmup || instance.config.warmup === 'none') return { error: 'null' };
let res;
if (typeof createImageBitmap === 'function') res = await warmupBitmap(instance);
else if (typeof Image !== 'undefined') res = await warmupCanvas(instance);
else res = await warmupNode(instance);
const t1 = now();
if (instance.config.debug) log('Warmup', instance.config.warmup, Math.round(t1 - t0), 'ms');
instance.emit('warmup');
return res;
}

5
test/test-main.js
View File

@ -66,7 +66,7 @@ async function testInstance(human) {
// if (!human.tf) human.tf = tf;
log('info', 'human version:', human.version);
log('info', 'platform:', human.sysinfo.platform, 'agent:', human.sysinfo.agent);
log('info', 'platform:', human.env.platform, 'agent:', human.env.agent);
log('info', 'tfjs version:', human.tf.version.tfjs);
await human.load();
@ -132,6 +132,7 @@ async function test(Human, inputConfig) {
}
const t0 = process.hrtime.bigint();
const human = new Human(config);
// await human.tf.ready();
await testInstance(human);
config.warmup = 'none';
await testWarmup(human, 'default');
@ -158,6 +159,8 @@ async function test(Human, inputConfig) {
testDetect(second, 'samples/ai-face.jpg', 'default'),
testDetect(human, 'samples/ai-body.jpg', 'default'),
testDetect(second, 'samples/ai-body.jpg', 'default'),
testDetect(human, 'samples/ai-upper.jpg', 'default'),
testDetect(second, 'samples/ai-upper.jpg', 'default'),
]);
const t1 = process.hrtime.bigint();
log('info', 'test complete:', Math.trunc(Number(t1 - t0) / 1000 / 1000), 'ms');

11
test/test-node-wasm.js
View File

@ -1,3 +1,5 @@
const tf = require('@tensorflow/tfjs/dist/tf.node.js'); // wasm backend requires tfjs to be loaded first
const wasm = require('@tensorflow/tfjs-backend-wasm/dist/tf-backend-wasm.node.js'); // wasm backend does not get auto-loaded in nodejs
const Human = require('../dist/human.node-wasm.js').default;
const test = require('./test-main.js').test;
@ -10,17 +12,20 @@ const config = {
async: false,
face: {
enabled: true,
detector: { enabled: true, rotation: true },
detector: { enabled: true, rotation: false },
mesh: { enabled: true },
iris: { enabled: true },
description: { enabled: true },
emotion: { enabled: true },
},
hand: { enabled: true },
hand: { enabled: true, rotation: false },
body: { enabled: true },
object: { enabled: false },
object: { enabled: true },
segmentation: { enabled: true },
filter: { enabled: false },
};
// @ts-ignore // in nodejs+wasm must set explicitly before using human
wasm.setWasmPaths(config.wasmPath); tf.setBackend('wasm');
test(Human, config);

2
test/test-node.js
View File

@ -14,7 +14,7 @@ const config = {
description: { enabled: true },
emotion: { enabled: true },
},
hand: { enabled: true },
hand: { enabled: true, rotation: true },
body: { enabled: true },
object: { enabled: true },
segmentation: { enabled: true },