tf.nn.atrous_conv2d

Atrous convolution (a.k.a. convolution with holes or dilated convolution).

This function is a simpler wrapper around the more general tf.nn.convolution, and exists only for backwards compatibility. You can use tf.nn.convolution to perform 1-D, 2-D, or 3-D atrous convolution.

Computes a 2-D atrous convolution, also known as convolution with holes or dilated convolution, given 4-D value and filters tensors. If the rate parameter is equal to one, it performs regular 2-D convolution. If the rate parameter is greater than one, it performs convolution with holes, sampling the input values every rate pixels in the height and width dimensions. This is equivalent to convolving the input with a set of upsampled filters, produced by inserting rate - 1 zeros between two consecutive values of the filters along the height and width dimensions, hence the name atrous convolution or convolution with holes (the French word trous means holes in English).

More specifically:

output[batch, height, width, out_channel] =
    sum_{dheight, dwidth, in_channel} (
        filters[dheight, dwidth, in_channel, out_channel] *
        value[batch, height + rate*dheight, width + rate*dwidth, in_channel]
    )

Atrous convolution allows us to explicitly control how densely to compute feature responses in fully convolutional networks. Used in conjunction with bilinear interpolation, it offers an alternative to conv2d_transpose in dense prediction tasks such as semantic image segmentation, optical flow computation, or depth estimation. It also allows us to effectively enlarge the field of view of filters without increasing the number of parameters or the amount of computation.

For a description of atrous convolution and how it can be used for dense feature extraction, please see: (Chen et al., 2015). The same operation is investigated further in (Yu et al., 2016). Previous works that effectively use atrous convolution in different ways are, among others, (Sermanet et al., 2014) and (Giusti et al., 2013). Atrous convolution is also closely related to the so-called noble identities in multi-rate signal processing.

There are many different ways to implement atrous convolution (see the refs above). The implementation here reduces

    atrous_conv2d(value, filters, rate, padding=padding)

to the following three operations: