You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I think the document of convolution is still quite confusing right now.
What does the different dimension of strides = [1, 1, 1, 1] mean?
Why the definition of "VALID" and "SAME" padding is quite different from the Stanford cs131n tutorial.
I think most of the people learn CNN according to the tutorials online. Is it possible to modify the document according to the tutorial? Thank you very much.
The text was updated successfully, but these errors were encountered:
NianzuMa
changed the title
Question about padding type definition
Question about the part of convolution document
Jan 2, 2016
NianzuMa
changed the title
Question about the part of convolution document
Question about document about convolution
Jan 2, 2016
NianzuMa
changed the title
Question about document about convolution
Question about the API document of convolution
Jan 2, 2016
The strides = [1, 2, 2, 1] are the strides for each of the dimension of the input tensor [batch, in_height, in_width, in_channels]. For the tf.nn.conv2d function the dimension 0 and 3 must have strides equal to 1. Based on what @Yangqing said in #196 I think the padding scheme is non-intuitive due to legacy reasons.
In issue 196, the problem of definition of different padding is mentioned.
https://www.tensorflow.org/versions/master/api_docs/python/nn.html#convolution
I think the document of convolution is still quite confusing right now.
What does the different dimension of
strides = [1, 1, 1, 1]
mean?Why the definition of "VALID" and "SAME" padding is quite different from the Stanford cs131n tutorial.
I think most of the people learn CNN according to the tutorials online. Is it possible to modify the document according to the tutorial? Thank you very much.
The text was updated successfully, but these errors were encountered: