How to do average subtraction and normalization with Tensorflow

Question

In http://cs231n.imtqy.com/neural-networks-2/ it is mentioned that for convolutional neural networks, it is preferable to project data using methods of subtraction and normalization of the average value.

I'm just wondering what is the best fit with Tensorflow.

Mean

X -= np.mean(X)

Normalization

X /= np.std(X, axis = 0)

Answer 1

You are looking for

• tf.image.per_image_whitening(image)if you are using a version of Tensorflow <r0.12
• tf.image.per_image_standardization(image) otherwise.

Linearly scale the image to have a zero average and unit rate.

op (x - mean)/adjust_stddev, - , adjust_stddev = max (stddev, 1.0/sqrt (image.NumElements())).

Answer 2

, real_valued_column , :

real_valued_column("col_name", normalizer = lambda x: (x-X.mean())/X.std())

X - . , , . .

Answer 3

- tf.image.per_image_standardization. https://www.tensorflow.org/api_docs/python/tf/image/per_image_standardization .