conv neural network - What does TensorFlow's `conv2d_transpose()` operation do? -
the documentation conv2d_transpose()
operation not explain does:
the transpose of conv2d.
this operation called "deconvolution" after deconvolutional networks, transpose (gradient) of conv2d rather actual deconvolution.
i went through paper doc points to, did not help.
what operation , examples of why want use it?
this best explanation i've seen online how convolution transpose works here.
i'll give own short description. applies convolution fractional stride. in other words spacing out input values (with zeroes) apply filter on region that's potentially smaller filter size.
as why 1 want use it. can used sort of upsampling learned weights opposed bilinear interpolation or other fixed form of upsampling.
Comments
Post a Comment