CODEWITHSUNDEEP
javascripttensorflowtensorflow.js
Józef Podlecki
Józef Podlecki

Reputation: 11283

Error when checking target: expected dense_Dense2 to have shape [,5], but got array with shape [5,1]

I was doing image classification tutorial on Tensorflow website and I can't make the section which does model.fit.

Here's the error.

Error: Error when checking target: expected dense_Dense2 to have shape [,5], but got array with shape [5,1].

It seems it's not happy about the model which I am passing.

I tried passing tf.tensor3D[] structure but it also didn't like it:

Error: Error when checking model input: the Array of Tensors that you are passing to your model is not the size the model expected. Expected to see 1 Tensor(s), but instead got the following list of Tensor(s): Tensor

As for now its [num_of_images, img_height, img_width, num_of_color_channels]

Cheers!

const button = document.getElementById("random");
const randomPokemon = document.getElementById("randomPokemon");
const prediction = document.getElementById("prediction");
const imageWidth = 128;
const imageHeight = 128;

const createModel = (classes) => {
    const model = tf.sequential();

    model.add(tf.layers.conv2d({
        inputShape: [imageHeight, imageWidth, 3],
        kernelSize: [3, 3],
        filters: 16,
        padding: "same",
        activation: 'relu'
    }));
    model.add(tf.layers.maxPooling2d({}));
    model.add(tf.layers.conv2d({
        kernelSize: [3, 3],
        filters: 32,
        padding: "same",
        activation: 'relu'
    }));
    model.add(tf.layers.maxPooling2d({}));
    model.add(tf.layers.conv2d({
        kernelSize: [3, 3],
        filters: 64,
        padding: "same",
        activation: 'relu'
    }));
    model.add(tf.layers.maxPooling2d({}));
    model.add(tf.layers.flatten({}));
    model.add(tf.layers.dense({units: 128, activation: 'relu'}));
    model.add(tf.layers.dense({ units: classes }));
    model.compile({
        optimizer: "adam",
        loss: "categoricalCrossentropy",
        metrics: ['accuracy'],
    });

    return model;
}

const getTensorFromImage = (src) => new Promise((resolve, reject) => {
    const image = new Image();
    
    image.onload = () => {
        const canvas = document.createElement('canvas');
        const context = canvas.getContext('2d');
        canvas.width = imageWidth;
        canvas.height = imageHeight;
        image.width = image.naturalWidth;
        image.height = image.naturalHeight;

        context.drawImage(image, 0, 0, imageWidth, imageHeight);
        
        const tensor = tf.browser.fromPixels(canvas);

        resolve(tensor);
    };
    image.crossOrigin = "";
    image.referrerPolicy = "origin"
    image.src = src;
});

const pokemons = [
    "wartortle.png",
    "bulbasaur.png",
    "charmander.png",
    "blastoise.png",
    "kakuna.png",
  ];
  
const getRandomPokemon = () => {
  const index = Math.floor(Math.random() * pokemons.length);
  const pokemon = pokemons[index];

  return `https://cors-anywhere.herokuapp.com/https://img.pokemondb.net/sprites/bank/normal/${pokemon}`
}

const model = createModel(pokemons.length);

(async () => {

const tensor_promises = pokemons.map(pr => getTensorFromImage(`https://cors-anywhere.herokuapp.com/https://img.pokemondb.net/sprites/bank/normal/${pr}`))
  const tensors = await Promise.all(tensor_promises);
  labels_normalized = pokemons.map((pr, index) => index);
  const label_tensor = tf.tensor1d(labels_normalized);
  
  const computedData = tensors.map(pr => pr.arraySync());
  const tensor4d = tf.tensor4d(computedData, [computedData.length, ...tensors[0].shape])

  await model.fit(tensor4d, label_tensor, {
    epochs: 20,
  })
})();

const onClick = async () => {
  const pokemonSrc = getRandomPokemon();
  randomPokemon.src = pokemonSrc;
  const tensor = await getTensorFromImage(pokemonSrc);
  
  const model_prediction = model.predict([tensor]);
  const predictions = model_prediction[0];
  const pokemon = pokemonss[predictions[0]];
  
  prediction.textContent = `Prediction: ${pokemon}`;
}

button.addEventListener("click", onClick);
<script src="https://cdnjs.cloudflare.com/ajax/libs/tensorflow/2.7.0/tf.min.js"></script>

<img id="randomPokemon"/>
<span id="prediction"></span>
<div>
  <button id="random">Random Pokemon</button>
</div>

Upvotes: 1

Views: 116

Answers (1)

edkeveked
edkeveked

Reputation: 18371

The tensor used for predictions should be 4d tensor. The shape of what is expected can be viewed as an array of what the inputShape is. So here it is an array of 3d tensors which therefore is a 4d tensor. If you only have one image (3d tensor), you can consider expanding the first axis (.expandDims(0)). That way, it will be as if you have an array of your image.

Upvotes: 1

Related Questions