CODEWITHSUNDEEP
pythontensorflowdeep-learningpadding
Yuki.U
Yuki.U

Reputation: 65

Does padding contain the number to use for calculation formula about new_height and new_width?

I'm trying to build the convolutional layers in TensorFlow. Actually it is for answering a quiz below. They give us the formula to achieve new_height and new_width too. But I don't know what is the number to put for padding because I thought padding='SAME' or padding='VALID' and it didn't contain number.

new_height = (input_height - filter_height + 2 * P)/S + 1

new_width = (input_width - filter_width + 2 * P)/S + 1 

# `tf.nn.conv2d` requires the input be 4D (batch_size, height, width, depth)
# (1, 4, 4, 1)
x = np.array([
    [0, 1, 0.5, 10],
    [2, 2.5, 1, -8],
    [4, 0, 5, 6],
    [15, 1, 2, 3]], dtype=np.float32).reshape((1, 4, 4, 1))
X = tf.constant(x)

def conv2d(input):
# Filter (weights and bias)
# The shape of the filter weight is (height, width, input_depth,    
   output_depth)
# The shape of the filter bias is (output_depth,)
# TODO: Define the filter weights `F_W` and filter bias `F_b`.
# NOTE: Remember to wrap them in `tf.Variable`, they are trainable 
        parameters after all.
F_W = ????
F_b = ????

# TODO: Set the stride for each dimension (batch_size, height, width, depth)
strides = ????
# TODO: set the padding, either 'VALID' or 'SAME'.
padding = ?

# `tf.nn.conv2d` does not include the bias computation so we have to add it   
ourselves after.

return tf.nn.conv2d(input, F_W, strides, padding) + F_b

out = conv2d(X)

Upvotes: 2

Views: 112

Answers (1)

Vipin Makde
Vipin Makde

Reputation: 86

SAME padding means the size of output feature-maps are the same as the input feature-maps (under the assumption of stride=1). For instance, if the input is Nin is of size 28×28, then in the output you expect to get Nout feature maps each of size 28×28 as well.

On the other hand, VALID padding means the size of the output feature-maps is reduced. For instance, if the input is Nin is of size 28x28, stride=1 and filter size is 3x3, then the output Nout feature-maps would be each of [{(28-3)/1}+1 = 26] 26x26.

To deduce the output the formula is: {(size of Input - size of Filter)/stride} + 1

hence from the above formula, you can see that the size of the output feature-map is inversely proportional to the stride (Higher the stride value, smaller will be the output feature map)

Upvotes: 2

Related Questions