From tensorflow\python\ops\gen_nn_ops.py
@tf_export('nn.conv2d')
def conv2d(input, filter, strides, padding, use_cudnn_on_gpu=True, data_format="NHWC", dilations=[1, 1, 1, 1], name=None):
r"""Computes a 2-D convolution given 4-D `input` and `filter` tensors.
给定4-D`input`和`filter`张量来计算2-D卷积。
Given an input tensor of shape `[batch, in_height, in_width, in_channels]`
and a filter / kernel tensor of shape
`[filter_height, filter_width, in_channels, out_channels]`, this op
performs the following:
给定形状为[[batch，in_height，in_width，in_channels]`的输入张量和形状为
[[filter_height，filter_width，in_channels，out_channels]`的过滤器/内核张量，
此操作执行以下操作：
1. Flattens the filter to a 2-D matrix with shape
`[filter_height * filter_width * in_channels, output_channels]`.
将过滤器展平为形状为[[filter_height * filter_width * in_channels，
output_channels]`的二维矩阵。
2. Extracts image patches from the input tensor to form a *virtual*
tensor of shape `[batch, out_height, out_width,
filter_height * filter_width * in_channels]`.
从输入张量中提取图像补丁，以形成形状为[[batch，out_height，out_width，
filter_height * filter_width * in_channels]`的*虚拟*张量。
3. For each patch, right-multiplies the filter matrix and the image patch
vector.
对于每个色块，将滤波器矩阵和图像色块向量右乘。
In detail, with the default NHWC format,
详细而言，使用默认的NHWC格式，
公式1：
output[b, i, j, k] =
sum_{di, dj, q} input[b, strides[1] * i + di, strides[2] * j + dj, q] *
filter[di, dj, q, k]
Must have `strides[0] = strides[3] = 1`. For the most common case of the same
horizontal and vertices strides, `strides = [1, stride, stride, 1]`.
必须具有“步幅[0] =步幅[3] = 1”。
对于相同的水平和顶点步幅，在最常见的情况下，“步幅= [1，步幅，步幅，1]”。
Args:
input: A `Tensor`. Must be one of the following types: `half`, `bfloat16`, `float32`, `float64`.
A 4-D tensor. The dimension order is interpreted according to the value
of `data_format`, see below for details.
张量 必须是以下类型之一：`half`，`bfloat16`，`float32`，`float64`。
一个4-D张量。 维度顺序是根据data_format的值来解释的，有关详细信息，请参见下文。
filter: A `Tensor`. Must have the same type as `input`.
A 4-D tensor of shape
`[filter_height, filter_width, in_channels, out_channels]`
张量，必须和输入类型相同。 形状为[[filter_height，filter_width，
in_channels，out_channels]`的4-D张量
strides: A list of `ints`.
1-D tensor of length 4. The stride of the sliding window for each
dimension of `input`. The dimension order is determined by the value of
`data_format`, see below for details.
`ints'的列表。 一维张量，长度为4。“ input”的每个维度的滑动窗口的步幅。
尺寸顺序由data_format的值确定，有关详细信息，请参见下文。
padding: A `string` from: `"SAME", "VALID"`.
The type of padding algorithm to use.
来自“ SAME”，“ VALID”`的`string`。 要使用的填充算法的类型。
use_cudnn_on_gpu: An optional `bool`. Defaults to `True`.
可选的布尔。 默认为True。
data_format: An optional `string` from: `"NHWC", "NCHW"`. Defaults to `"NHWC"`.
Specify the data format of the input and output data. With the
default format "NHWC", the data is stored in the order of:
[batch, height, width, channels].
Alternatively, the format could be "NCHW", the data storage order of:
[batch, channels, height, width].
可选字符串，来自“ NHWC”，“ NCHW”。 默认为““ NHWC”`。
指定输入和输出数据的数据格式。
使用默认格式“ NHWC”时，数据按以下顺序存储：[批，高度，宽度，通道]。
或者，格式可以是“ NCHW”，数据存储顺序为：[批，通道，高度，宽度]。
dilations: An optional list of `ints`. Defaults to `[1, 1, 1, 1]`.
1-D tensor of length 4. The dilation factor for each dimension of
`input`. If set to k > 1, there will be k-1 skipped cells between each
filter element on that dimension. The dimension order is determined by the
value of `data_format`, see above for details. Dilations in the batch and
depth dimensions must be 1.
`ints'的可选列表。 默认为[[1，1，1，1]`。 长度为1的一维张量。
“输入”的每个维度的膨胀因子。 如果设置为k> 1，则该维度上每个过滤器元素之间将有k-1个跳过的像元。
尺寸顺序由data_format的值确定，有关详细信息，请参见上文。
批次尺寸和深度尺寸的膨胀必须为1。
name: A name for the operation (optional).
操作的名称（可选）。
Returns:
A `Tensor`. Has the same type as `input`.
张量 与`input`具有相同的类型。
"""
_ctx = _context._context
if _ctx is None or not _ctx._eager_context.is_eager:
if not isinstance(strides, (list, tuple)):
raise TypeError(
"Expected list for 'strides' argument to "
"'conv2d' Op, not %r." % strides)
strides = [_execute.make_int(_i, "strides") for _i in strides]
padding = _execute.make_str(padding, "padding")
if use_cudnn_on_gpu is None:
use_cudnn_on_gpu = True
use_cudnn_on_gpu = _execute.make_bool(use_cudnn_on_gpu, "use_cudnn_on_gpu")
if data_format is None:
data_format = "NHWC"
data_format = _execute.make_str(data_format, "data_format")
if dilations is None:
dilations = [1, 1, 1, 1]
if not isinstance(dilations, (list, tuple)):
raise TypeError(
"Expected list for 'dilations' argument to "
"'conv2d' Op, not %r." % dilations)
dilations = [_execute.make_int(_i, "dilations") for _i in dilations]
_, _, _op = _op_def_lib._apply_op_helper(
"Conv2D", input=input, filter=filter, strides=strides,
padding=padding, use_cudnn_on_gpu=use_cudnn_on_gpu,
data_format=data_format, dilations=dilations, name=name)
_result = _op.outputs[:]
_inputs_flat = _op.inputs
_attrs = ("T", _op.get_attr("T"), "strides", _op.get_attr("strides"),
"use_cudnn_on_gpu", _op.get_attr("use_cudnn_on_gpu"), "padding",
_op.get_attr("padding"), "data_format",
_op.get_attr("data_format"), "dilations",
_op.get_attr("dilations"))
_execute.record_gradient(
"Conv2D", _inputs_flat, _attrs, _result, name)
_result, = _result
return _result
else:
try:
_result = _pywrap_tensorflow.TFE_Py_FastPathExecute(
_ctx._context_handle, _ctx._eager_context.device_name, "Conv2D", name,
_ctx._post_execution_callbacks, input, filter, "strides", strides,
"use_cudnn_on_gpu", use_cudnn_on_gpu, "padding", padding,
"data_format", data_format, "dilations", dilations)
return _result
except _core._FallbackException:
return conv2d_eager_fallback(
input, filter, strides=strides, use_cudnn_on_gpu=use_cudnn_on_gpu,
padding=padding, data_format=data_format, dilations=dilations,
name=name, ctx=_ctx)
except _core._NotOkStatusException as e:
if name is not None:
message = e.message + " name: " + name
else:
message = e.message
_six.raise_from(_core._status_to_exception(e.code, message), None)
Official 1:
