Node.js环境下TensorFlow.js报错：Cannot read property 'backend' of undefined

依赖项 #

"dependencies": {
    "@tensorflow/tfjs-backend-webgl": "^4.6.0",
    "@tensorflow/tfjs-core": "^4.6.0",
    "@tensorflow/tfjs-node": "^4.6.0",
    "canvas": "^2.11.2",
    "express": "^4.18.2",
    "fs": "^0.0.1-security",
    "jsdom": "^21.1.2",
    "node-cmd": "^5.0.0",
    "node-gyp": "^9.3.1",
    "path": "^0.12.7",
    "png-js": "^1.0.0",
    "xml-js": "^1.6.11"
}

代码执行步骤 #

1. 加载模块

const tf = require('@tensorflow/tfjs-node');
require('@tensorflow/tfjs-backend-webgl');

2. 加载模型

const model = await tf.loadGraphModel(path);

3. 图像预处理

const batched = tf.tidy(() => {
        const imageTensor = tf.browser.fromPixels(cnv);
        return (
          imageTensor
            .resizeBilinear([width, height])
            .expandDims(0)
            .toInt()
        );
});

4. 执行模型

return model.executeAsync(batched,["StatefulPartitionedCall/Postprocessor/Slice","StatefulPartitionedCall/Postprocessor/ExpandDims_1"]).then(function (result){
    console.log("flag 0");
} // 从未执行到该日志

报错信息 #

@tensorflow\tfjs-core\dist\tf-core.node.js:4519
    var srcBackend = info.backend;
                          ^

TypeError: Cannot read properties of undefined (reading 'backend')
    at Engine.moveData (\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:4519:31)
    at DataStorage.get (\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:230:28)
    at NodeJSKernelBackend.getInputTensorIds (\node_modules\@tensorflow\tfjs-node\dist\nodejs_kernel_backend.js:169:43)
    at NodeJSKernelBackend.executeSingleOutput (\node_modules\@tensorflow\tfjs-node\dist\nodejs_kernel_backend.js:216:73)
    at Object.kernelFunc (\node_modules\@tensorflow\tfjs-node\dist\kernels\Reshape.js:34:27)
    at kernelFunc (\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:4704:32)
    at \node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:4764:27
    at Engine.scopedRun (\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:4569:23)
    at Engine.runKernelFunc (\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:4760:14)
    at Engine.runKernel (\node_modules\@tensorflow\tfjs-core\dist\tf-core.node.js:4633:21)

运行环境 #

• Windows Anaconda环境
• Python 2.7（TensorFlow.js推荐版本）
• Node.js v18.15.0

问题判断 #

已尝试所有能找到的解决方案，仍无法解决，怀疑是TensorFlow.js的已知bug：未定义后端（tfjs-node）#7706

解决建议 #

1. 移除冲突后端依赖：Node环境下使用tfjs-node时，无需手动引入tfjs-backend-webgl，两者后端会冲突，直接删除require('@tensorflow/tfjs-backend-webgl');这一行。

2. 替换Node环境的图像读取API：tf.browser.fromPixels是浏览器专属API，Node环境下改用tf.node.fromPixels结合canvas库读取图像，避免跨环境API导致的张量后端异常，示例：

const { createCanvas, loadImage } = require('canvas');
const img = await loadImage(imagePath);
const canvas = createCanvas(img.width, img.height);
const ctx = canvas.getContext('2d');
ctx.drawImage(img, 0, 0);
const imageTensor = tf.node.fromPixels(canvas);

3. 检查张量生命周期管理：避免tf.tidy不当使用导致的张量后端信息丢失，可尝试手动构建张量替代tidy包裹的逻辑。

4. 调整版本兼容性：Node.js v18与TensorFlow.js 4.6.0可能存在兼容问题，可尝试降级Node.js到v16，或升级TensorFlow.js至最新稳定版。

内容的提问来源于stack exchange，提问作者Typo