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
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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/operators/roi_pool_op.h"
namespace paddle {
namespace operators {
class ROIPoolOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
void InferShape(framework::InferShapeContext* ctx) const override {
PADDLE_ENFORCE(ctx->HasInput("X"),
"Input(X) of ROIPoolOp should not be null.");
PADDLE_ENFORCE(ctx->HasInput("ROIs"),
"Input(ROIs) of ROIPoolOp should not be null.");
PADDLE_ENFORCE(ctx->HasOutput("Out"),
"Output(Out) of ROIPoolOp should not be null.");
PADDLE_ENFORCE(ctx->HasOutput("Argmax"),
"Output(Argmax) of ROIPoolOp should not be null.");
auto input_dims = ctx->GetInputDim("X");
auto rois_dims = ctx->GetInputDim("ROIs");
PADDLE_ENFORCE(input_dims.size() == 4,
"The format of input tensor is NCHW.");
PADDLE_ENFORCE(rois_dims.size() == 2,
"ROIs should be a 2-D tensor of shape (num_rois, 5)"
"given as [[batch_id, x1, y1, x2, y2], …].");
int pooled_height = ctx->Attrs().Get<int>("pooled_height");
int pooled_width = ctx->Attrs().Get<int>("pooled_width");
float spatial_scale = ctx->Attrs().Get<float>("spatial_scale");
PADDLE_ENFORCE_GT(pooled_height, 0,
"The pooled output height must greater than 0");
PADDLE_ENFORCE_GT(pooled_width, 0,
"The pooled output width must greater than 0");
PADDLE_ENFORCE_GT(spatial_scale, 0.0f,
"The spatial scale must greater than 0");
auto out_dims = input_dims;
out_dims[0] = rois_dims[0];
out_dims[1] = input_dims[1];
out_dims[2] = pooled_height;
out_dims[3] = pooled_width;
ctx->SetOutputDim("Out", out_dims);
ctx->SetOutputDim("Argmax", out_dims);
}
protected:
framework::OpKernelType GetKernelType(
const framework::ExecutionContext& ctx) const override {
return framework::OpKernelType(
framework::ToDataType(ctx.Input<framework::Tensor>("X")->type()),
ctx.device_context());
}
};
class ROIPoolGradOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
void InferShape(framework::InferShapeContext* ctx) const override {
PADDLE_ENFORCE(ctx->HasInput(framework::GradVarName("Out")),
"The gradient of Out should not be null.");
PADDLE_ENFORCE(ctx->HasOutputs(framework::GradVarName("X")),
"The gradient of X should not be null.");
ctx->SetOutputsDim(framework::GradVarName("X"), ctx->GetInputsDim("X"));
}
protected:
framework::OpKernelType GetKernelType(
const framework::ExecutionContext& ctx) const override {
return framework::OpKernelType(
framework::ToDataType(ctx.Input<framework::Tensor>("X")->type()),
ctx.device_context());
}
};
class ROIPoolOpMaker : public framework::OpProtoAndCheckerMaker {
public:
ROIPoolOpMaker(framework::OpProto* proto,
framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddInput("X",
"(Tensor), "
"the input of ROIPoolOp. "
"The format of input tensor is NCHW. Where N is batch size, "
"C is the number of input channels, "
"H is the height of the feature, and "
"W is the width of the feature.");
AddInput("ROIs",
"(Tensor), "
"ROIs (Regions of Interest) to pool over. "
"should be a 2-D tensor of shape (num_rois, 5)"
"given as [[batch_id, x1, y1, x2, y2], …]. "
"Where batch_id is the id of the data, "
"(x1, y1) is the top left coordinates, and "
"(x2, y2) is the bottom right coordinates.");
AddOutput("Out",
"(Tensor), "
"The output of ROIPoolOp is a 4-D tensor with shape "
"(num_rois, channels, pooled_h, pooled_w).");
AddOutput("Argmax",
"(Tensor), "
"Argmaxes corresponding to indices in X used "
"for gradient computation. Only output "
"if arg “is_test” is false.").AsIntermediate();
AddAttr<float>("spatial_scale",
"(float, default 1.0), "
"Multiplicative spatial scale factor "
"to translate ROI coords from their input scale "
"to the scale used when pooling.")
.SetDefault(1.0);
AddAttr<int>("pooled_height",
"(int, default 1), "
"The pooled output height.")
.SetDefault(1);
AddAttr<int>("pooled_width",
"(int, default 1), "
"The pooled output width.")
.SetDefault(1);
AddComment(R"DOC(
ROIPool operator
ROI Pooling for Faster-RCNN. The link below is a further introduction:
https://stackoverflow.com/questions/43430056/what-is-roi-layer-in-fast-rcnn
)DOC");
}
};
} // namespace operators
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(roi_pool, ops::ROIPoolOp, ops::ROIPoolOpMaker,
roi_pool_grad, ops::ROIPoolGradOp);
REGISTER_OP_CPU_KERNEL(
roi_pool,
ops::CPUROIPoolOpKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(
roi_pool_grad,
ops::CPUROIPoolGradOpKernel<paddle::platform::CPUPlace, float>);