如何解决：ValueError：检查输入时出错：期望flatten_input具有3个维，但数组的形状为（28，28）

Question

I'm trying to input my own image into the mnist model

mnist = tf.keras.datasets.mnist

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
model.fit(x_train, y_train, epochs=5)

model.evaluate(x_test, y_test)
model.predict(gray)

gray is an image that is of shape (28,28) but I am getting an error stating that the model expects 3 dimensions even though the input shape is (28,28).

The code works if I do gray.reshape(1,28,28) but I don't know why that works or if that is even the correct solution to this problem.

Answer 1

The model instance expects a batch of images. This is specified on this line:

tf.keras.layers.Flatten(input_shape=(28, 28))

When you specify the input_shape=(28, 28) , you are basically telling Tensorflow that you will receive a batch of inputs, where each element in the batch will have shape 28 x 28 . So, when you add your image, make sure to expand its dimensions:

gray = np.expand_dims(gray, axis=0)

Then, you can safely do:

model.predict(gray)

---

Besides, in this particular case you are fine using np.reshape . However, that method serves a different purpose so I would stick to np.expand_dims . Here is proof to see that they are equal:

X = np.random.rand(28, 28)
np.testing.assert_array_equal(np.expand_dims(X, axis=0), np.reshape(X (1, 28, 28)))
# The assert passes