CODEWITHSUNDEEP
pythontensorflowmnist
Maarten Faddegon
Maarten Faddegon

Reputation: 785

How to get predictions for one image from the MNIST with TensorFlow?

I followed this tutorial https://www.tensorflow.org/tutorials/layers and trained a model for recognizing hand written numbers from the MNIST set.

The following code works as expected and prints for each image in the set the probability and class

mnist = tf.contrib.learn.datasets.load_dataset("mnist")
train_data = mnist.train.images  # Returns np.array
tf.reset_default_graph()  
with tf.Session() as sess:

  mnist_classifier = tf.estimator.Estimator(model_fn=cnn_model_fn, model_dir="model/")

  pred = mnist_classifier.predict(input_fn=tf.estimator.inputs.numpy_input_fn(
      x={"x": train_data},
      shuffle=False))

  for p in pred:
    print(p)

However, when I instead try to predict for just one image with

mnist_classifier.predict(input_fn=tf.estimator.inputs.numpy_input_fn(
          x={"x": train_data[0]},
          shuffle=False))

My program fails and TensorFlow reports

InvalidArgumentError: Input to reshape is a tensor with 128 values,
but the requested shape requires a multiple of 784

This puzzles me because when I print the length of the first image from the set it reports 784

print("length of input: {}".format(len(train_data[0]))

How do I get the predictions for just one image?

Upvotes: 2

Views: 2113

Answers (2)

Domi W
Domi W

Reputation: 646

You can also use tf.expand_dims. The documentation says:

This operation is useful if you want to add a batch dimension to a single element. For example, if you have a single image of shape [height, width, channels], you can make it a batch of one image with expand_dims(image, 0), which will make the shape [1, height, width, channels].

Upvotes: 0

xdurch0
xdurch0

Reputation: 10474

This is most likely related to the fact that you are dropping the batch dimension when creating the single-item dataset. What I mean by that is that you should use

mnist_classifier.predict(input_fn=tf.estimator.inputs.numpy_input_fn(
      x={"x": np.array([train_data[0])]},
      shuffle=False))

instead. Note the additional list wrapped around train_data[0]. This will take the array of shape [1, 784] and create a dataset with one element, this in turn being a vector with 784 elements. As your code is right now, you are basically creating a dataset with 784 elements, each of which is a single number. This leads to shape mismatches down the road.

Upvotes: 1

Related Questions