tensorflow.js部署到web 笔记

BUG：

无法加载源映射

Could not read source map for https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@latest: Unexpected 404 response from https://cdn.jsdelivr.net/npm/@tensorflow/tf.min.js.map: Failed to resolve the requested file.

解决：（https://www.nuomiphp.com/eplan/188052.html）

换成"<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs/dist/tf.min.js"> </script>"

加载模型报错 lambda，inception_resnet_v2报错，inceptionV3并不报错

Uncaught (in promise) Error: Unknown layer: Lambda. This may be due to one of the following reasons:

The layer is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
The custom layer is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().
at new t (errors.ts:48)
at deserializeKerasObject (generic_utils.ts:202)
at deserialize (serialization.ts:27)
at i (container.ts:1329)
at t.fromConfig (container.ts:1355)
at deserializeKerasObject (generic_utils.ts:235)
at deserialize (serialization.ts:27)
at models.ts:270
at index.ts:79
at Object.next (index.ts:79)

参考：

https://github.com/tensorspace-team/tensorspace/issues/121

https://stackoverflow.com/questions/66616597/tensorflow-js-error-in-loading-augmentation-layers-operation

https://www.jianshu.com/p/dd2bebfa867d

https://zhuanlan.zhihu.com/p/86355195

bug2在替换掉lambda后出现，基础库

Uncaught (in promise) TypeError: undefined is not iterable (cannot read property Symbol(Symbol.iterator))
    at Fv (runtime.js:747)
    at t.fromConfig (container.js:1232)
    at eO (generic_utils.js:278)
    at aM (serialization.js:31)
    at u (container.js:1206)
    at t.fromConfig (container.js:1234)
    at eO (generic_utils.js:278)
    at aM (serialization.js:31)
    at models.js:295
    at c (runtime.js:63)

tf.fromPixels
```
Uncaught (in promise) TypeError: tf.fromPixels is not a function
```
已解决：https://stackoverflow.com/questions/55402551/why-am-i-getting-the-error-tf-frompixels-is-not-a-function

tf.fromPixels was deprecated in version 1.0.0, use: tf.browser.fromPixels()

getImageData

'getImageData' on 'CanvasRenderingContext2D': Value is not of type 'long'.

我是这么写的：

const imgData = context.getImageData(0,0,canvas.width,canvas.height);

已解决：换成const imgData = context.getImageData(0,0,299,299);

tensorflow.js没有tf.argmax，变成了tf.argMax

加载模型时路径的写法

const my_model = await tf.loadLayersModel("model_js/model.json")
//正确

  const my_model = await tf.loadLayersModel("http://127.0.0.1:5502/TensorFlow%20Deployment/Course%201%20-%20TensorFlow-JS/Week%204/Examples/model_js/model.json")
//正确,前面一大串是工作区文件夹名称

tensorflow.js的预测结果与tensorflow大相径庭

图片：hym_003(-)_180.jpg

tensorflow.js

[[0.0038602, 0.0000056, 0.136271, 0.0231699, 4e-7, 5e-7, 1e-7, 0.0000079, 4e-7, 0.836684],]

tensorflow-python

array([[6.1449555e-08, 7.2196451e-07, 9.9997318e-01, 2.7251541e-08,
        4.4864832e-09, 2.8673458e-07, 1.8537765e-05, 2.0686712e-06,
        9.0471025e-10, 5.0913218e-06]], dtype=float32)

https://github.com/tensorflow/tfjs/issues/776

后续更新部署。