What is the difference between xavier_initializer and xavier_initializer_conv2d?

Question

I noticed that TensorFlow 1.0 contains two Xavier initialization assistants in contrib,

• tf.contrib.layers.xavier_initializer () and
• tf.contrib.layers.xavier_initializer_conv2d ()

Both link to the same page of documentation and have the same signature:

tf.contrib.layers.xavier_initializer(uniform=True, seed=None, dtype=tf.float32)
tf.contrib.layers.xavier_initializer_conv2d(uniform=True, seed=None, dtype=tf.float32)

however, the difference between them is not explained at all. I can guess by name that the version _conv2dshould be used for 2D convolutional layers, but will it have a noticeable effect if you use the regular version?

Answer 1

. :

(2010): . .

, . : x = sqrt(6. / (in + out)); [-x, x] sqrt(3. / (in + out)).