tf.conv | TensorFlow v2.16.1 (original) (raw)
tf.conv
Stay organized with collections Save and categorize content based on your preferences.
Computes a N-D convolution given (N+1+batch_dims)-D input and (N+2)-D filter tensors.
View aliases
Compat aliases for migration
SeeMigration guide for more details.
tf.conv(
input: Annotated[Any, TV_Conv_T],
filter: Annotated[Any, TV_Conv_T],
strides,
padding: str,
explicit_paddings=[],
data_format: str = 'CHANNELS_LAST',
dilations=[],
batch_dims: int = 1,
groups: int = 1,
name=None
) -> Annotated[Any, TV_Conv_T]
General function for computing a N-D convolution. It is required that1 <= N <= 3.
| Args | |
|---|---|
| input | A Tensor. Must be one of the following types: half, bfloat16, float32, float64, int32. Tensor of type T and shape batch_shape + spatial_shape + [in_channels] in the case that channels_last_format = true or shapebatch_shape + [in_channels] + spatial_shape if channels_last_format = false. spatial_shape is N-dimensional with N=2 or N=3. Also note that batch_shape is dictated by the parameter batch_dimsand defaults to 1. |
| filter | A Tensor. Must have the same type as input. An (N+2)-D Tensor with the same type as input and shapespatial_filter_shape + [in_channels, out_channels], where spatial_filter_shape is N-dimensional with N=2 or N=3. |
| strides | A list of ints. 1-D tensor of length N+2. The stride of the sliding window for each dimension of input. Must have strides[0] = strides[N+1] = 1. |
| padding | A string from: "SAME", "VALID", "EXPLICIT". The type of padding algorithm to use. |
| explicit_paddings | An optional list of ints. Defaults to []. If padding is "EXPLICIT", the list of explicit padding amounts. For the ith dimension, the amount of padding inserted before and after the dimension isexplicit_paddings[2 * i] and explicit_paddings[2 * i + 1], respectively. Ifpadding is not "EXPLICIT", explicit_paddings must be empty. |
| data_format | An optional string from: "CHANNELS_FIRST", "CHANNELS_LAST". Defaults to "CHANNELS_LAST". Used to set the data format. By default CHANNELS_FIRST, uses NHWC (2D) / NDHWC (3D) or if CHANNELS_LAST, uses NCHW (2D) / NCDHW (3D). |
| dilations | An optional list of ints. Defaults to []. 1-D tensor of length N+2. The dilation factor for each dimension ofinput. 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 channels_last_format, see above for details. Dilations in the batch and depth dimensions must be 1. |
| batch_dims | An optional int. Defaults to 1. A positive integer specifying the number of batch dimensions for the input tensor. Should be less than the rank of the input tensor. |
| groups | An optional int. Defaults to 1. A positive integer specifying the number of groups in which the input is split along the channel axis. Each group is convolved separately withfilters / groups filters. The output is the concatenation of all the groups results along the channel axis. Input channels and filters must both be divisible by groups. |
| name | A name for the operation (optional). |
| Returns |
|---|
| A Tensor. Has the same type as input. |
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates. Some content is licensed under the numpy license.
Last updated 2024-04-26 UTC.