CODEWITHSUNDEEP
pythontensorflowdeep-learningconv-neural-network
Devin Haslam
Devin Haslam

Reputation: 759

Tensorflow: Preparing input data

I am attempting to replicate a deep neural network from a research paper. The architecture can be found here:

I have completed designing the model, and now I am attempting to prepare training data. I have been using the tensorflow tutorials found here as a guide: https://www.tensorflow.org/get_started/mnist/pros

In the case of the mnist data, a 27x27 image is converted to a 1d vector for x. On the other hand, y_ has the shape [none, 10] because each image has the possibility to labeled 10 different ways (0-9)

 x = tf.placeholder(tf.float32, shape=[None, 784])
 y_ = tf.placeholder(tf.float32, shape=[None, 10])

My data is a 32x32x7 3d image so x is easy to calculate. 

 x = tf.placeholder(tf.float32, shape=[None, 7168])

Although my image is 32x32x7, each pixel has a density and label associated with it. I believe the density values will be loaded into x and the labels would be loaded into y. Is this a correct assumption or should I be loading my data in a different way? 

 y_ = tf.placeholder(tf.float32, shape=[None, 7168])

Upvotes: 1

Views: 525

Answers (1)

musically_ut
musically_ut

Reputation: 34288

my image is 32x32x7, each pixel has a density and label associated with it

If so, then the output of the network, and the target y_, would be of shape: 

[
 None,         # Batch size
 32 * 32 * 7,  # Vector size
 N             # N target labels (one hot encoded)
]

Upvotes: 1

Related Questions