The transpose of `conv3d`. This operation is sometimes called "deconvolution" after [Deconvolutional Networks](https://www.matthewzeiler.com/mattzeiler/deconvolutionalnetworks.pdf), but is really the transpose (gradient) of `conv3d` rather than an actual deconvolution. Args: value: A
(
value,
filter=None, # pylint: disable=redefined-builtin
output_shape=None,
strides=None,
padding="SAME",
data_format="NDHWC",
name=None,
input=None, # pylint: disable=redefined-builtin
filters=None,
dilations=None)
| 2510 | |
| 2511 | @tf_export(v1=["nn.conv3d_transpose"]) |
| 2512 | def conv3d_transpose( |
| 2513 | value, |
| 2514 | filter=None, # pylint: disable=redefined-builtin |
| 2515 | output_shape=None, |
| 2516 | strides=None, |
| 2517 | padding="SAME", |
| 2518 | data_format="NDHWC", |
| 2519 | name=None, |
| 2520 | input=None, # pylint: disable=redefined-builtin |
| 2521 | filters=None, |
| 2522 | dilations=None): |
| 2523 | """The transpose of `conv3d`. |
| 2524 | |
| 2525 | This operation is sometimes called "deconvolution" after [Deconvolutional |
| 2526 | Networks](https://www.matthewzeiler.com/mattzeiler/deconvolutionalnetworks.pdf), |
| 2527 | but is really the transpose (gradient) of `conv3d` rather than an actual |
| 2528 | deconvolution. |
| 2529 | |
| 2530 | Args: |
| 2531 | value: A 5-D `Tensor` of type `float` and shape |
| 2532 | `[batch, depth, height, width, in_channels]`. |
| 2533 | filter: A 5-D `Tensor` with the same type as `value` and shape |
| 2534 | `[depth, height, width, output_channels, in_channels]`. `filter`'s |
| 2535 | `in_channels` dimension must match that of `value`. |
| 2536 | output_shape: A 1-D `Tensor` representing the output shape of the |
| 2537 | deconvolution op. |
| 2538 | strides: A list of ints. The stride of the sliding window for each |
| 2539 | dimension of the input tensor. |
| 2540 | padding: A string, either `'VALID'` or `'SAME'`. The padding algorithm. |
| 2541 | See the "returns" section of `tf.nn.convolution` for details. |
| 2542 | data_format: A string, either `'NDHWC'` or `'NCDHW`' specifying the layout |
| 2543 | of the input and output tensors. Defaults to `'NDHWC'`. |
| 2544 | name: Optional name for the returned tensor. |
| 2545 | input: Alias of value. |
| 2546 | filters: Alias of filter. |
| 2547 | dilations: An int or list of `ints` that has length `1`, `3` or `5`, |
| 2548 | defaults to 1. The dilation factor for each dimension of`input`. If a |
| 2549 | single value is given it is replicated in the `D`, `H` and `W` dimension. |
| 2550 | By default the `N` and `C` dimensions are set to 1. If set to k > 1, there |
| 2551 | will be k-1 skipped cells between each filter element on that dimension. |
| 2552 | The dimension order is determined by the value of `data_format`, see above |
| 2553 | for details. Dilations in the batch and depth dimensions if a 5-d tensor |
| 2554 | must be 1. |
| 2555 | |
| 2556 | Returns: |
| 2557 | A `Tensor` with the same type as `value`. |
| 2558 | |
| 2559 | Raises: |
| 2560 | ValueError: If input/output depth does not match `filter`'s shape, or if |
| 2561 | padding is other than `'VALID'` or `'SAME'`. |
| 2562 | """ |
| 2563 | filter = deprecated_argument_lookup("filters", filters, "filter", filter) |
| 2564 | value = deprecated_argument_lookup("input", input, "value", value) |
| 2565 | return conv3d_transpose_v2( |
| 2566 | value, |
| 2567 | filter, |
| 2568 | output_shape, |
| 2569 | strides, |
nothing calls this directly
no test coverage detected