transpose_convolution.hpp
10 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#ifndef OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_TRANSPOSE_CONVOLUTION_HPP
#define OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_TRANSPOSE_CONVOLUTION_HPP
#include "../../op_cuda.hpp"
#include "../csl/cudnn.hpp"
#include "../csl/stream.hpp"
#include "../csl/tensor.hpp"
#include "../csl/tensor_ops.hpp"
#include "../kernels/scale_shift.hpp"
#include <opencv2/core.hpp>
#include <cstddef>
#include <cstdint>
#include <vector>
#include <utility>
#include <algorithm>
namespace cv { namespace dnn { namespace cuda4dnn {
struct TransposeConvolutionConfiguration {
/* other than `input_shape` and `output_shape`, all the configuration values must be provided
* for the corresponding convolution operation (not transpose convolution)
*/
/* the size of the following vectors must be equal to the kernel size */
std::vector<std::size_t> kernel_size;
std::vector<std::size_t> dilations, strides;
enum class PaddingMode {
MANUAL, /* uses explicit padding values provided in `pads_begin` and `pads_end` */
VALID, /* no padding is added */
SAME /* TensorFlow logic is used for same padding */
};
/* explicit paddings are used if and only if padMode is set to manual */
PaddingMode padMode;
std::vector<std::size_t> pads_begin, pads_end;
/* full shape inclusive of channel and batch axis */
std::vector<std::size_t> input_shape;
std::vector<std::size_t> output_shape;
/* group count for grouped convolution */
std::size_t groups;
};
template <class T>
class TransposeConvolutionOp final : public CUDABackendNode {
public:
using wrapper_type = GetCUDABackendWrapperType<T>;
TransposeConvolutionOp(csl::Stream stream_, csl::cudnn::Handle handle, const TransposeConvolutionConfiguration& config, const Mat& filters, const Mat& bias)
: stream(std::move(stream_)), cudnnHandle(std::move(handle))
{
/* we make use of backward pass of convolution to perform forward pass of transpose convolution
* hence, we must setup configuration for the convolution operation and perform backward pass
*/
const auto& kernel_size = config.kernel_size;
const auto& dilations = config.dilations;
const auto& strides = config.strides;
const auto convolution_order = kernel_size.size();
CV_Assert(convolution_order >= 1);
CV_Assert(convolution_order == dilations.size());
CV_Assert(convolution_order == strides.size());
const auto& input_shape = config.input_shape;
const auto& output_shape = config.output_shape;
CV_Assert(input_shape.size() == output_shape.size());
CV_Assert(input_shape.size() == convolution_order + 2);
const auto groups = config.groups;
if (convolution_order > 3)
CV_Error(Error::StsNotImplemented, "Only 1D/2D/3D transpose convolution is supported.");
const auto rank = input_shape.size();
const auto input_feature_maps = input_shape[1];
const auto output_feature_maps = output_shape[1];
const auto output_feature_maps_per_group = output_feature_maps / groups;
CV_Assert(output_feature_maps % groups == 0);
filtersTensor = csl::makeTensorHeader<T>(filters);
csl::copyMatToTensor<T>(filters, filtersTensor, stream);
if (!bias.empty())
{
CV_Assert(bias.total() == output_feature_maps);
biasTensor = csl::makeTensorHeader<T>(bias);
csl::copyMatToTensor<T>(bias, biasTensor, stream);
}
/* left and right are misleading as the padding is applicable for any number of dimensions
* but we use those identifiers to avoid confusion with `pads_begin` and `pads_end`
*
* `common_padding` contains the amount of padding that has to be added to both sides
* `padding_left` and `padding_right` contains the amount of padding that needs to be added
* to a particular side in addition to the common padding
*
* note that we compute the padding for the convolution operation
*/
std::vector<std::size_t> common_padding(rank, 0);
std::vector<std::size_t> padding_left(rank, 0), padding_right(rank, 0);
if (config.padMode == TransposeConvolutionConfiguration::PaddingMode::MANUAL)
{
const auto& pads_begin = config.pads_begin;
const auto& pads_end = config.pads_end;
CV_Assert(convolution_order == pads_begin.size());
CV_Assert(convolution_order == pads_end.size());
for (int i = 2; i < common_padding.size(); i++)
{
common_padding[i] = std::min(pads_begin[i - 2], pads_end[i - 2]);
padding_left[i] = pads_begin[i - 2] - common_padding[i];
padding_right[i] = pads_end[i - 2] - common_padding[i];
}
}
else if (config.padMode == TransposeConvolutionConfiguration::PaddingMode::VALID)
{
/* nothing to do as the paddings are already preset to zero */
}
else if (config.padMode == TransposeConvolutionConfiguration::PaddingMode::SAME)
{
/* TensorFlow Logic:
* total_padding[i] = (o[i] - 1) * s[i] + effective_k[i] - i[i]
*
* if total padding is odd, the extra is added towards the end
*/
for (int i = 2; i < rank; i++)
{
const auto j = i - 2; /* filter index */
const auto effective_kernel_size = dilations[j] * (kernel_size[j] - 1) + 1;
const auto required_total_padding =
std::max<std::int64_t>(0, (input_shape[i] - 1) * strides[j] + effective_kernel_size - output_shape[i]);
common_padding[i] = required_total_padding / 2;
padding_left[i] = 0;
padding_right[i] = required_total_padding % 2;
}
}
/* in some scenarios, the extra padding at the end may not change the output at all */
for (int i = 2; i < rank; i++) {
const auto j = i - 2; /* filter idx */
const auto total_padding = common_padding[i] * 2 + padding_left[i] + padding_right[i];
const auto effective_kernel_size = dilations[j] * (kernel_size[j] - 1) + 1;
std::int64_t rem = (input_shape[i] + total_padding - effective_kernel_size) % strides[j];
/* the output shape doesn't change if we decrease the total padding by at most `rem`
* provided that we decrease from the right
*/
if (rem && padding_right[i] > 0)
padding_right[i] = std::max<std::int64_t>(0, padding_right[i] - rem);
}
auto is_not_zero = [](std::size_t i) { return i != 0; };
if(std::any_of(std::begin(padding_left), std::end(padding_left), is_not_zero) ||
std::any_of(std::begin(padding_right), std::end(padding_right), is_not_zero))
{
CV_Error(Error::StsNotImplemented, "Padding configuration requires asymmetric padding and hence is not supported.");
}
typename csl::TransposeConvolution<T>::params_type params;
params.input_shape.assign(std::begin(input_shape), std::end(input_shape));
params.output_shape.assign(std::begin(output_shape), std::end(output_shape));
auto& fshape = params.filter_shape;
fshape.resize(rank);
fshape[0] = input_feature_maps;
fshape[1] = output_feature_maps_per_group;
std::copy(std::begin(kernel_size), std::end(kernel_size), std::begin(fshape) + 2);
CV_Assert(fshape.size() == kernel_size.size() + 2);
params.padding.assign(std::begin(common_padding) + 2, std::end(common_padding));
params.stride = strides;
params.dilation = dilations;
params.groups = config.groups;
convoluter = csl::TransposeConvolution<T>(cudnnHandle, params);
csl::WorkspaceBuilder builder;
builder.require(convoluter.get_workspace_size());
scratch_mem_in_bytes = builder.required_workspace_size();
}
void forward(
const std::vector<cv::Ptr<BackendWrapper>>& inputs,
const std::vector<cv::Ptr<BackendWrapper>>& outputs,
csl::Workspace& workspace) override
{
CV_Assert(inputs.size() == 1 && outputs.size() == 1);
auto input_wrapper = inputs[0].dynamicCast<wrapper_type>();
auto input = input_wrapper->getView();
auto output_wrapper = outputs[0].dynamicCast<wrapper_type>();
auto output = output_wrapper->getSpan();
csl::WorkspaceAllocator allocator(workspace);
convoluter.transpose_convolve(output, input, filtersTensor, allocator.get_instance());
if (!biasTensor.empty())
{
std::size_t inner_size = total(output_wrapper->getShape(), 2, -1);
kernels::biasN<T>(stream, output, output, inner_size, biasTensor);
}
}
std::size_t get_workspace_memory_in_bytes() const noexcept override { return scratch_mem_in_bytes; }
private:
csl::Stream stream;
csl::cudnn::Handle cudnnHandle;
csl::Tensor<T> filtersTensor, biasTensor;
csl::TransposeConvolution<T> convoluter;
std::size_t scratch_mem_in_bytes;
};
}}} /* namespace cv::dnn::cuda4dnn */
#endif /* OPENCV_DNN_SRC_CUDA4DNN_PRIMITIVES_TRANSPOSE_CONVOLUTION_HPP */