我正在使用 tensorflow bodypix 模型创建一个身体分割应用程序.它在浏览器中运行良好.我正在使用 webpack 来使用它的模块(见下文)

import * as wasm from "@tensorflow/tfjs-backend-wasm";从@tensorflow/tfjs-core"导入 * as tf；从@tensorflow-models/body-pix"导入*作为bodyPix；wasm.setWasmPaths("./wasm/");tf.setBackend("wasm").then(() => {//一些简单的香草js代码});//更多的香草js代码...

它在 chrome 中运行良好，运行 npx webpack 后按预期输出.

但是，当简单地通过创建一个主电子文件来使用电子运行它时，它只会输出一个空白的白色屏幕，并在控制台中显示以下错误-

Uncaught TypeError: this.util.TextEncoder 不是构造函数在新的<匿名>(main.js:2)

它指向的行来自一个看起来像这样的缩小代码-

...SOME_CODE...&&Me().setPlatform("node",new class{构造函数(){this.util=n(758),this.textEncoder=new this.util.TextEncoder}...SOME_MORE_CODE...

我认为电子只是没有顶栏的铬，但这似乎是错误的.有人可以帮我吗我正在使用以下版本-

<块引用>

nodejs v12.16.3"、electron11.1.1"、tfjs2.8.2"

查看chrome和electron的截图-

IN CHROME(点击放大)

...................................................

IN ELECTRON(点击放大)

解决方案

解决方案

我以前有

wasm.setWasmPaths("./wasm/");tf.setBackend("wasm").then(() => {//一些简单的香草js代码});

在我的主代码中，我已将文件夹从 wasm(dist/) 复制到项目的文件夹中.

从我的项目文件夹中删除相同的内容并将代码更改为 -

wasm.setWasmPaths("../node_modules/@tensorflow/tfjs-backend-wasm/dist/");//或从 ./开始，如果您的主文件与 node_modules 位于同一文件夹中tf.setBackend("wasm").then(() => {//...});

我是怎么回到这里的?

首先感谢@edkeveked 的努力并指出我

在 Electron App (Nodejs) 中加载 TensorflowJS 时出错p>

我通过创建一个 electron hello world 项目然后添加 tfjs 和 tfjs-backend-wasm 得到了解决方案.新项目工作正常，但是即使将 node_modules 从新项目移动到旧项目也不适用于旧项目.但是一旦我更改了 wasm 路径，它就可以正常工作.

更新:

现在我已经多次遇到这个问题，每次都是通过创建一个新文件夹来解决的，首先安装电子并先创建一个简单的电子应用程序，然后安装其他依赖项并将旧代码复制到新文件夹中.(警告:不要'不复制节点模块文件夹)

好像是tfjs或者electron的bug

I am creating a body segmentation app using tensorflow bodypix model. It works fine in the browser. I am using webpack to use its modules(see below)

import * as wasm from "@tensorflow/tfjs-backend-wasm";
import * as tf from "@tensorflow/tfjs-core";
import * as bodyPix from "@tensorflow-models/body-pix";

wasm.setWasmPaths("./wasm/");
tf.setBackend("wasm").then(() => {
  //some simple vanilla js code
});
//some more vanilla js code...

It works exactly fine in chrome and giving output as expected after running npx webpack .

However when irun it with electron simply by creating a main electron file it outputs nothing but a blank white screen with the following error in console-

Uncaught TypeError: this.util.TextEncoder is not a constructor
    at new <anonymous> (main.js:2)

the line where it is pointing is from a minified codew which looks like this-

...SOME_CODE...&&Me().setPlatform("node",new class{
constructor(){this.util=n(758),this.textEncoder=new this.util.TextEncoder}...SOME_MORE_CODE...

i thought that electron is simply chrome without top bars, but this seems wrong. can someone help me here i am using following versions-

"nodejs v12.16.3", "electron11.1.1", "tfjs2.8.2"

see the screen shot of chrome and electron-

IN CHROME(click to enlarge)

................................................

IN ELECTRON(click to enlarge)

解决方案

THE SOLUTION

i previously have

wasm.setWasmPaths("./wasm/");
tf.setBackend("wasm").then(() => {
  //some simple vanilla js code
});

in my main code, and i have copied the folder from wasm(dist/) to project's folder.

Deleting the same from my project's folder and changing the code to -

wasm.setWasmPaths("../node_modules/@tensorflow/tfjs-backend-wasm/dist/"); //or start from ./ if your main file is in same folder as node_modules
tf.setBackend("wasm").then(() => {
  //...
});

How i recahed here?

at first thanks to @edkeveked for his effort and pointing me to

Error loading TensorflowJS in Electron App (Nodejs)

i got the solution by creating an electron hello world project and then adding tfjs, then tfjs-backend-wasm. the new project is working correctly but however even moving the node_modules from new project to older one is not working for the older. but as soon i changed the wasm path, it worked giving no error.

Update:

now I have encountered the problem several times and everytime it's solved by creating a new folder, first installing electron and creating a simple electron app first, and then installing other dependencies and copying old code in the new folder.(warning: don't copy the node modules folder)

It seems to be a bug in tfjs or electron