unpooling.cc 8.2 KB
Newer Older
1
/* Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
S
sweetsky0901 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14

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. */

Y
Yi Wang 已提交
15
#include "paddle/fluid/operators/math/unpooling.h"
16

S
sweetsky0901 已提交
17 18 19
namespace paddle {
namespace operators {
namespace math {
S
sweetsky0901 已提交
20
template <typename T>
Q
QI JUN 已提交
21
class Unpool2dMaxFunctor<platform::CPUDeviceContext, T> {
S
sweetsky0901 已提交
22
 public:
Q
QI JUN 已提交
23
  void operator()(const platform::CPUDeviceContext& context,
S
sweetsky0901 已提交
24 25
                  const framework::Tensor& input,
                  const framework::Tensor& indices, framework::Tensor* output) {
S
sweetsky0901 已提交
26 27 28 29 30 31 32 33 34
    const int batch_size = input.dims()[0];
    const int input_height = input.dims()[2];
    const int input_width = input.dims()[3];
    const int output_channels = output->dims()[1];
    const int output_height = output->dims()[2];
    const int output_width = output->dims()[3];
    int input_feasize = input_height * input_width;
    int output_feasize = output_height * output_width;
    const T* input_data = input.data<T>();
S
sweetsky0901 已提交
35
    const int* indices_data = indices.data<int>();
S
sweetsky0901 已提交
36 37 38 39
    T* output_data = output->mutable_data<T>(context.GetPlace());
    for (int b = 0; b < batch_size; ++b) {
      for (int c = 0; c < output_channels; ++c) {
        for (int i = 0; i < input_feasize; ++i) {
S
sweetsky0901 已提交
40
          int index = indices_data[i];
41

42 43 44 45 46 47 48
          PADDLE_ENFORCE_LT(
              index, output_feasize,
              platform::errors::InvalidArgument(
                  "index should less than output tensor height * output tensor "
                  "width. Expected %ld < %ld, but got "
                  "%ld >= %ld. Please check input value.",
                  index, output_feasize, index, output_feasize));
S
sweetsky0901 已提交
49 50 51 52 53 54 55 56 57
          output_data[index] = input_data[i];
        }
        input_data += input_feasize;
        indices_data += input_feasize;
        output_data += output_feasize;
      }
    }
  }
};
S
sweetsky0901 已提交
58
template <class T>
Q
QI JUN 已提交
59
class Unpool2dMaxGradFunctor<platform::CPUDeviceContext, T> {
S
sweetsky0901 已提交
60
 public:
Q
QI JUN 已提交
61
  void operator()(const platform::CPUDeviceContext& context,
S
sweetsky0901 已提交
62 63 64 65 66
                  const framework::Tensor& input,
                  const framework::Tensor& indices,
                  const framework::Tensor& output,
                  const framework::Tensor& output_grad,
                  framework::Tensor* input_grad) {
S
sweetsky0901 已提交
67 68 69
    const int batch_size = input.dims()[0];
    const int input_height = input.dims()[2];
    const int input_width = input.dims()[3];
S
sweetsky0901 已提交
70 71 72
    const int output_channels = output.dims()[1];
    const int output_height = output.dims()[2];
    const int output_width = output.dims()[3];
S
sweetsky0901 已提交
73 74
    int input_feasize = input_height * input_width;
    int output_feasize = output_height * output_width;
S
sweetsky0901 已提交
75
    const int* indices_data = indices.data<int>();
S
sweetsky0901 已提交
76 77 78 79 80
    const T* output_grad_data = output_grad.data<T>();
    T* input_grad_data = input_grad->mutable_data<T>(context.GetPlace());

    for (int b = 0; b < batch_size; ++b) {
      for (int c = 0; c < output_channels; ++c) {
S
sweetsky0901 已提交
81
        for (int i = 0; i < input_feasize; ++i) {
S
sweetsky0901 已提交
82
          int index = indices_data[i];
83 84 85 86 87 88 89
          PADDLE_ENFORCE_LT(
              index, output_feasize,
              platform::errors::InvalidArgument(
                  "index should less than output tensor height * output tensor "
                  "width. Expected %ld < %ld, but got "
                  "%ld >= %ld. Please check input value.",
                  index, output_feasize, index, output_feasize));
S
sweetsky0901 已提交
90 91 92 93 94 95 96 97 98
          input_grad_data[i] = output_grad_data[index];
        }
        input_grad_data += input_feasize;
        indices_data += input_feasize;
        output_grad_data += output_feasize;
      }
    }
  }
};
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

template <typename T>
class Unpool3dMaxFunctor<platform::CPUDeviceContext, T> {
 public:
  void operator()(const platform::CPUDeviceContext& context,
                  const framework::Tensor& input,
                  const framework::Tensor& indices, framework::Tensor* output) {
    const int batch_size = input.dims()[0];
    const int input_depth = input.dims()[2];
    const int input_height = input.dims()[3];
    const int input_width = input.dims()[4];
    const int output_channels = output->dims()[1];
    const int output_depth = output->dims()[2];
    const int output_height = output->dims()[3];
    const int output_width = output->dims()[4];
    int input_feasize = input_depth * input_height * input_width;
    int output_feasize = output_depth * output_height * output_width;
    const T* input_data = input.data<T>();
    const int* indices_data = indices.data<int>();
    T* output_data = output->mutable_data<T>(context.GetPlace());
    for (int b = 0; b < batch_size; ++b) {
      for (int c = 0; c < output_channels; ++c) {
        for (int i = 0; i < input_feasize; ++i) {
          int index = indices_data[i];

          PADDLE_ENFORCE_LT(
              index, output_feasize,
              platform::errors::InvalidArgument(
                  "index should less than output tensor depth * output tensor "
                  "height "
                  "* output tensor width. Expected %ld < %ld, but got "
                  "%ld >= %ld. Please check input value.",
                  index, output_feasize, index, output_feasize));
          output_data[index] = input_data[i];
        }
        input_data += input_feasize;
        indices_data += input_feasize;
        output_data += output_feasize;
      }
    }
  }
};
template <class T>
class Unpool3dMaxGradFunctor<platform::CPUDeviceContext, T> {
 public:
  void operator()(const platform::CPUDeviceContext& context,
                  const framework::Tensor& input,
                  const framework::Tensor& indices,
                  const framework::Tensor& output,
                  const framework::Tensor& output_grad,
                  framework::Tensor* input_grad) {
    const int batch_size = input.dims()[0];
    const int input_depth = input.dims()[2];
    const int input_height = input.dims()[3];
    const int input_width = input.dims()[4];
    const int output_channels = output.dims()[1];
    const int output_depth = output.dims()[2];
    const int output_height = output.dims()[3];
    const int output_width = output.dims()[4];
    int input_feasize = input_depth * input_height * input_width;
    int output_feasize = output_depth * output_height * output_width;
    const int* indices_data = indices.data<int>();
    const T* output_grad_data = output_grad.data<T>();
    T* input_grad_data = input_grad->mutable_data<T>(context.GetPlace());

    for (int b = 0; b < batch_size; ++b) {
      for (int c = 0; c < output_channels; ++c) {
        for (int i = 0; i < input_feasize; ++i) {
          int index = indices_data[i];
          PADDLE_ENFORCE_LT(
              index, output_feasize,
              platform::errors::InvalidArgument(
                  "index should less than output tensor depth * output tensor "
                  "height "
                  "* output tensor width. Expected %ld < %ld, but got "
                  "%ld >= %ld. Please check input value.",
                  index, output_feasize, index, output_feasize));
          input_grad_data[i] = output_grad_data[index];
        }
        input_grad_data += input_feasize;
        indices_data += input_feasize;
        output_grad_data += output_feasize;
      }
    }
  }
};

Q
QI JUN 已提交
186 187 188 189
template class Unpool2dMaxGradFunctor<platform::CPUDeviceContext, float>;
template class Unpool2dMaxGradFunctor<platform::CPUDeviceContext, double>;
template class Unpool2dMaxFunctor<platform::CPUDeviceContext, float>;
template class Unpool2dMaxFunctor<platform::CPUDeviceContext, double>;
190 191 192 193
template class Unpool3dMaxGradFunctor<platform::CPUDeviceContext, float>;
template class Unpool3dMaxGradFunctor<platform::CPUDeviceContext, double>;
template class Unpool3dMaxFunctor<platform::CPUDeviceContext, float>;
template class Unpool3dMaxFunctor<platform::CPUDeviceContext, double>;
S
sweetsky0901 已提交
194 195 196
}  // namespace math
}  // namespace operators
}  // namespace paddle