forked from PaddlePaddle/Paddle
/
pixel_unshuffle_op.cc
130 lines (109 loc) · 4.94 KB
/
pixel_unshuffle_op.cc
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
/*Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/fluid/framework/infershape_utils.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/op_version_registry.h"
#include "paddle/phi/core/infermeta_utils.h"
#include "paddle/phi/infermeta/unary.h"
namespace paddle {
namespace operators {
class PixelUnshuffleOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
};
class PixelUnshuffleOpMaker : public framework::OpProtoAndCheckerMaker {
public:
void Make() override {
AddInput("X",
"(Tensor, default Tensor<float>), "
"the input feature data of PixelUnshuffleOp, the layout is "
"[N, C, H, W] or [N, H, W, C].");
AddOutput("Out",
"(Tensor, default Tensor<float>), the output of "
"PixelUnshuffleOp. The layout is [N, C*factor^2, H/factor, "
"W/factor] or [N, H/factor, W/factor, C*factor^2].");
AddAttr<int>("downscale_factor",
"the factor to decrease spatial resolution by.")
.SetDefault(1);
AddAttr<std::string>(
"data_format",
"An optional string from: \"NHWC\", \"NCHW\". "
"Defaults to \"NHWC\", Specify the data format of the input data.")
.SetDefault("NCHW");
AddComment(R"DOC(
Pixel Unshuffle operator
This operator rearranges elements in a tensor of shape :math:`(*, C, H, W)`
to a tensor of shape :math:`(*, C\times r^2, H / r, W / r)`.
This operation is the reversion of PixelShuffle operation.
Please refer to the paper:
`Real-Time Single Image and Video Super-Resolution Using an Efficient
Sub-Pixel Convolutional Neural Network <https://arxiv.org/abs/1609.05158v2>`_
by Shi et. al (2016) for more details.
)DOC");
}
};
template <typename T>
class PixelUnshuffleGradOpMaker : public framework::SingleGradOpMaker<T> {
public:
using framework::SingleGradOpMaker<T>::SingleGradOpMaker;
protected:
void Apply(GradOpPtr<T> op) const override {
op->SetType("pixel_unshuffle_grad");
op->SetInput(framework::GradVarName("Out"), this->OutputGrad("Out"));
op->SetAttrMap(this->Attrs());
op->SetOutput(framework::GradVarName("X"), this->InputGrad("X"));
}
};
class PixelUnshuffleGradOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
void InferShape(framework::InferShapeContext* ctx) const override {
PADDLE_ENFORCE_EQ(
ctx->HasInput(framework::GradVarName("Out")), true,
platform::errors::NotFound("Input(Out@Grad) should not be null"));
PADDLE_ENFORCE_EQ(
ctx->HasOutput(framework::GradVarName("X")), true,
platform::errors::NotFound("Output(X@Grad) should not be null"));
auto do_dims = ctx->GetInputDim(framework::GradVarName("Out"));
PADDLE_ENFORCE_EQ(do_dims.size(), 4,
platform::errors::InvalidArgument(
"Input should be a 4-D tensor of format [N, C, H, W] "
"or [N, H, W, C], but got %u.",
do_dims.size()));
auto downscale_factor = ctx->Attrs().Get<int>("downscale_factor");
const std::string data_format =
ctx->Attrs().Get<std::string>("data_format");
const bool channel_last = (data_format == "NHWC");
auto dx_dims = do_dims;
dx_dims[0] = do_dims[0];
if (!channel_last) {
dx_dims[1] = do_dims[1] / (downscale_factor * downscale_factor);
dx_dims[2] = do_dims[2] * downscale_factor;
dx_dims[3] = do_dims[3] * downscale_factor;
} else {
dx_dims[1] = do_dims[1] * downscale_factor;
dx_dims[2] = do_dims[2] * downscale_factor;
dx_dims[3] = do_dims[3] / (downscale_factor * downscale_factor);
}
ctx->SetOutputDim(framework::GradVarName("X"), dx_dims);
}
};
} // namespace operators
} // namespace paddle
namespace ops = paddle::operators;
DECLARE_INFER_SHAPE_FUNCTOR(pixel_unshuffle, PixelUnshuffleInferShapeFunctor,
PD_INFER_META(phi::PixelUnshuffleInferMeta));
REGISTER_OPERATOR(pixel_unshuffle, ops::PixelUnshuffleOp,
ops::PixelUnshuffleOpMaker,
ops::PixelUnshuffleGradOpMaker<paddle::framework::OpDesc>,
ops::PixelUnshuffleGradOpMaker<paddle::imperative::OpBase>,
PixelUnshuffleInferShapeFunctor);
REGISTER_OPERATOR(pixel_unshuffle_grad, ops::PixelUnshuffleGradOp);