Computes sums of N-D convolutions (actually cross correlation). It is required that 1 <= N <= 3. This is used to implement the more generic `convolution` function, which extends the interface of this function with a `dilation_rate` parameter. Args: input: Rank N+2 tensor of type T of
(
input, # pylint: disable=redefined-builtin
filter, # pylint: disable=redefined-builtin
padding,
data_format=None, # pylint: disable=redefined-builtin
strides=None,
name=None)
| 84 | |
| 85 | |
| 86 | def _non_atrous_convolution( |
| 87 | input, # pylint: disable=redefined-builtin |
| 88 | filter, # pylint: disable=redefined-builtin |
| 89 | padding, |
| 90 | data_format=None, # pylint: disable=redefined-builtin |
| 91 | strides=None, |
| 92 | name=None): |
| 93 | """Computes sums of N-D convolutions (actually cross correlation). |
| 94 | |
| 95 | It is required that 1 <= N <= 3. |
| 96 | |
| 97 | This is used to implement the more generic `convolution` function, which |
| 98 | extends the interface of this function with a `dilation_rate` parameter. |
| 99 | |
| 100 | Args: |
| 101 | |
| 102 | input: Rank N+2 tensor of type T of shape |
| 103 | `[batch_size] + input_spatial_shape + [in_channels]` if `data_format` |
| 104 | does not start with `"NC"`, or |
| 105 | `[batch_size, in_channels] + input_spatial_shape` if `data_format` starts |
| 106 | with `"NC"`. |
| 107 | filter: Rank N+2 tensor of type T of shape |
| 108 | `filter_spatial_shape + [in_channels, out_channels]`. Rank of either |
| 109 | `input` or `filter` must be known. |
| 110 | padding: Padding method to use, must be either "VALID" or "SAME". |
| 111 | data_format: A string or None. Specifies whether the channel dimension of |
| 112 | the `input` and output is the last dimension (default, or if `data_format` |
| 113 | does not start with "NC"), or the second dimension (if `data_format` |
| 114 | starts with "NC"). For N=1, the valid values are "NWC" (default) and |
| 115 | "NCW". For N=2, the valid values are "NHWC" (default) and "NCHW". |
| 116 | For N=3, the valid values are "NDHWC" (default) and "NCDHW". |
| 117 | strides: Sequence of N positive integers, defaults to `[1] * N`. |
| 118 | name: Name prefix to use. |
| 119 | |
| 120 | Returns: |
| 121 | Rank N+2 tensor of type T of shape |
| 122 | `[batch_size] + output_spatial_shape + [out_channels]`, where |
| 123 | if padding == "SAME": |
| 124 | output_spatial_shape = input_spatial_shape |
| 125 | if padding == "VALID": |
| 126 | output_spatial_shape = input_spatial_shape - filter_spatial_shape + 1. |
| 127 | |
| 128 | Raises: |
| 129 | ValueError: if ranks are incompatible. |
| 130 | |
| 131 | """ |
| 132 | with ops.name_scope(name, "non_atrous_convolution", [input, filter]) as scope: |
| 133 | input = ops.convert_to_tensor(input, name="input") # pylint: disable=redefined-builtin |
| 134 | input_shape = input.get_shape() |
| 135 | filter = ops.convert_to_tensor(filter, name="filter") # pylint: disable=redefined-builtin |
| 136 | filter_shape = filter.get_shape() |
| 137 | op = _Convolution( |
| 138 | input_shape, |
| 139 | filter_shape=filter_shape, |
| 140 | padding=padding, |
| 141 | data_format=data_format, |
| 142 | strides=strides, |
| 143 | name=scope) |
nothing calls this directly
no test coverage detected