Compute Library
 21.02
NEDeconvolutionLayer.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2021 Arm Limited.
3  *
4  * SPDX-License-Identifier: MIT
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to
8  * deal in the Software without restriction, including without limitation the
9  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10  * sell copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in all
14  * copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
24 #ifndef ARM_COMPUTE_NEDECONVOLUTIONLAYER_H
25 #define ARM_COMPUTE_NEDECONVOLUTIONLAYER_H
26 
31 
32 #include "arm_compute/core/Types.h"
37 
38 #include <memory>
39 
40 namespace arm_compute
41 {
42 /** Function to run the deconvolution layer.
43  *
44  * Deconvolution Layer is the backward pass of Convolution Layer. First we transform the input depending on the stride and pad info and then perfrom a 1x1
45  * convolution pass. Input stride defines how many zeroes we should put between each element of the input, pad is the amount of padding and finaly a is a user
46  * specified value where a < stride - 1 that increases the padding top and right of the input image.
47  *
48  * The relation between input to output is as follows:
49  * \f[
50  * width\_output = (width\_input - 1) \cdot stride\_x - 2 \cdot padding\_x + kernel\_x
51  * \f]
52  * \f[
53  * height\_output = (height\_input - 1) \cdot stride\_y - 2 \cdot padding\_y + kernel\_y
54  * \f]
55  *
56  * where
57  * width is the size of the first input dimension.
58  * height is the size of the second input dimension.
59  * width_output is the size of the first output dimension.
60  * height_output is the size of the second output dimension.
61  * kernel_x and kernel_y are the convolution sizes in x and y.
62  * stride_x and stride_y is the input stride of the first and second dimension.
63  *
64  * The weights used by Deconvolution are supposed to be the same as the ones used for Convolution. Therefore, it will be necessary to use the weights in the
65  * reverse order to perform an actual convolution. This is achieved by using @ref NEReverse.
66  *
67  * This function calls the following Neon kernels/functions:
68  *
69  * -# @ref CPPUpsample
70  * -# @ref NEConvolutionLayer
71  * -# @ref NEPermute
72  * -# @ref NEReverse
73  *
74  */
76 {
77 public:
78  /** Constructor */
79  NEDeconvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
80 
81  /** Prevent instances of this class from being copied (As this class contains pointers) */
83  /** Prevent instances of this class from being copied (As this class contains pointers) */
85  /** Prevent instances of this class from being moved (As this class contains pointers) */
87  /** Prevent instances of this class from being moved (As this class contains pointers) */
89  /** Default destructor */
90  virtual ~NEDeconvolutionLayer() = default;
91 
92  /** Set the input, weights, biases and output tensors.
93  *
94  * @param[in,out] input Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED.
95  * @param[in] weights The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input.
96  * @param[in] bias Optional, ignored if NULL. The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 and QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input.
97  * @param[out] output Output tensor. The output has the same number of dimensions as the @p input.
98  * @param[in] info Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
99  *
100  */
101  void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info);
102  /** Static function to check if given info will lead to a valid configuration of @ref NEDeconvolutionLayer
103  *
104  * @param[in] input Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED.
105  * @param[in] weights The 4d weights info with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input.
106  * @param[in] bias (Optional) The biases have one dimension. Data type supported: Data types supported: S32 for QASYMM8 and QASYMM8_SIGNED input, F32 for F32 input, F16 for F16 input.
107  * @param[in] output Output tensor info. The output has the same number of dimensions as the @p input.
108  * @param[in] info Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
109  *
110  * @return a status
111  */
112  static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &info);
113 
114  // Inherited methods overridden:
115  void run() override;
116  void prepare() override;
117 
118 private:
119  MemoryGroup _memory_group;
120  NEConvolutionLayer _conv_f;
121  CPPUpsample _upsample_f;
122  NEReverse _flip_weights;
123  Tensor _scaled_output;
124  Tensor _weights_flipped;
125  Tensor _flip_axis;
126  const ITensor *_original_weights;
127  ITensor *_input;
128  PadStrideInfo _info;
129  bool _is_prepared;
130 };
131 } // arm_compute
132 #endif /* ARM_COMPUTE_NEDECONVOLUTIONLAYER_H */
Base class for all functions.
Definition: IFunction.h:30
Function to run the deconvolution layer.
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &info)
Static function to check if given info will lead to a valid configuration of NEDeconvolutionLayer.
NEDeconvolutionLayer & operator=(const NEDeconvolutionLayer &)=delete
Prevent instances of this class from being copied (As this class contains pointers) ...
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
Status class.
Definition: Error.h:52
Interface for Neon tensor.
Definition: ITensor.h:36
void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info)
Set the input, weights, biases and output tensors.
Copyright (c) 2017-2021 Arm Limited.
void run() override
Run the kernels contained in the function.
Basic function to simulate a convolution layer.
Basic function to run CPPUpsample.
Definition: CPPUpsample.h:36
Basic implementation of the tensor interface.
Definition: Tensor.h:37
Padding and stride information class.
Definition: Types.h:722
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
NEDeconvolutionLayer(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Constructor.
void prepare() override
Prepare the function for executing.
virtual ~NEDeconvolutionLayer()=default
Default destructor.
Basic function to run NEReverseKernel.
Definition: NEReverse.h:37