有没有办法将预期的 conv2d_Conv2D1_input 从 4 维减少到 3 维？

Question

问题:

• a ValueError 是说 conv2d_Conv2D1_input 期望有 4 个维度，但是得到了形状为 [475,475,3]
的数组

但是：

• 输入形状设置为[475,475,3]
• 记录时，张量的形状为 [475,475,3]

错误：ValueError: Error when checking : expected conv2d_Conv2D1_input to have 4 dimension(s), but got array with shape [475,475,3]

张量：

Tensor {
  kept: false,
  isDisposedInternal: false,
  shape: [ 475, 475, 3 ],
  dtype: 'int32',
  size: 676875,
  strides: [ 1425, 3 ],
  dataId: {},
  id: 8,
  rankType: '3',
  scopeId: 4
}

完整代码：

var tf = require('@tensorflow/tfjs');
var tfnode = require('@tensorflow/tfjs-node');
var fs = require(`fs`)

const main = async () => {
  const loadImage = async (file) => {
    const imageBuffer = await fs.readFileSync(file)
    const tensorFeature = await tfnode.node.decodeImage(imageBuffer, 3)
    return tensorFeature;
  }

  const tensorFeature = await loadImage(`./1.png`)
  const tensorFeature2 = await loadImage(`./4.png`)
  const tensorFeature3 = await loadImage(`./7.png`)

  console.log(tensorFeature)
  console.log(tensorFeature2)
  console.log(tensorFeature3)

  tensorFeatures = [tensorFeature, tensorFeature2, tensorFeature3]

  labelArray = [0, 1, 2]
  tensorLabels = tf.oneHot(tf.tensor1d(labelArray, 'int32'), 3);

  const model = tf.sequential();
  model.add(tf.layers.conv2d({
    inputShape: [475, 475, 3],
    filters: 32,
    kernelSize: 3,
    activation: 'relu',
  }));
  model.add(tf.layers.flatten());
  model.add(tf.layers.dense({units: 3, activation: 'softmax'}));

  model.compile({
    optimizer: 'sgd',
    loss: 'categoricalCrossentropy',
    metrics: ['accuracy']
  });

  model.summary()

  model.fit(tf.stack(tensorFeatures), tensorLabels)

  const im = await loadImage(`./2.png`)
  model.predict(im)
}
main()

Answer 1

批量维度是使命。可以使用 expandDims()

添加

const im = await loadImage(`./2.png`).expandDims()
model.predict(im)