Open3D (C++ API)  0.15.1
ContinuousConvTransposeBackpropFilterOpKernel.h
Go to the documentation of this file.
1 // ----------------------------------------------------------------------------
2 // - Open3D: www.open3d.org -
3 // ----------------------------------------------------------------------------
4 // The MIT License (MIT)
5 //
6 // Copyright (c) 2018-2021 www.open3d.org
7 //
8 // Permission is hereby granted, free of charge, to any person obtaining a copy
9 // of this software and associated documentation files (the "Software"), to deal
10 // in the Software without restriction, including without limitation the rights
11 // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 // copies of the Software, and to permit persons to whom the Software is
13 // furnished to do so, subject to the following conditions:
14 //
15 // The above copyright notice and this permission notice shall be included in
16 // all copies or substantial portions of the Software.
17 //
18 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21 // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23 // FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 // IN THE SOFTWARE.
25 // ----------------------------------------------------------------------------
26 
27 #pragma once
28 
30 #include "tensorflow/core/framework/op.h"
31 #include "tensorflow/core/framework/op_kernel.h"
32 #include "tensorflow/core/lib/core/errors.h"
33 
34 template <class TIndex>
36  : public tensorflow::OpKernel {
37 public:
39  tensorflow::OpKernelConstruction* construction)
40  : OpKernel(construction) {
41  using namespace tensorflow;
42  using namespace open3d::ml::impl;
43  OP_REQUIRES_OK(construction,
44  construction->GetAttr("align_corners", &align_corners));
45  OP_REQUIRES_OK(construction,
46  construction->GetAttr("normalize", &normalize));
47 
48  std::string interpolation_str;
49  OP_REQUIRES_OK(construction, construction->GetAttr("interpolation",
50  &interpolation_str));
51 
52  if (interpolation_str == "linear")
53  interpolation = InterpolationMode::LINEAR;
54  else if (interpolation_str == "linear_border")
55  interpolation = InterpolationMode::LINEAR_BORDER;
56  else
58 
59  std::string mapping_str;
60  OP_REQUIRES_OK(construction, construction->GetAttr("coordinate_mapping",
61  &mapping_str));
62 
63  if (mapping_str == "ball_to_cube_radial")
64  coordinate_mapping = CoordinateMapping::BALL_TO_CUBE_RADIAL;
65  else if (mapping_str == "ball_to_cube_volume_preserving")
67  CoordinateMapping::BALL_TO_CUBE_VOLUME_PRESERVING;
68  else
69  coordinate_mapping = CoordinateMapping::IDENTITY;
70 
71  OP_REQUIRES_OK(construction, construction->GetAttr("max_temp_mem_MB",
72  &max_temp_mem_MB));
73  }
74 
75  void Compute(tensorflow::OpKernelContext* context) override {
76  using namespace tensorflow;
77  static_assert(sizeof(int64) == sizeof(int64_t),
78  "int64 type is not compatible");
79  const Tensor& filter = context->input(0);
80 
81  const Tensor& out_positions = context->input(1);
82  OP_REQUIRES(context,
83  out_positions.shape().dim_size(0) <=
84  std::numeric_limits<TIndex>::max(),
85  errors::InvalidArgument("Too many output points"));
86 
87  const Tensor& out_importance = context->input(2);
88  OP_REQUIRES(context,
89  out_importance.shape().dim_size(0) == 0 ||
90  out_importance.shape().dim_size(0) ==
91  out_positions.shape().dim_size(0),
92  errors::InvalidArgument("length of out_importance must "
93  "match the number of output points "
94  "or must be 0"));
95 
96  const Tensor& extents = context->input(3);
97 
98  const Tensor& offset = context->input(4);
99  OP_REQUIRES(context, offset.shape().dims() == 1,
100  errors::InvalidArgument("offset must be a rank 1 tensor"));
101  OP_REQUIRES(context, offset.shape().dim_size(0) == 3,
102  errors::InvalidArgument("offset length must be 3"));
103 
104  const Tensor& inp_positions = context->input(5);
105  OP_REQUIRES(context,
106  inp_positions.shape().dim_size(0) <=
107  std::numeric_limits<TIndex>::max(),
108  errors::InvalidArgument("Too many input points"));
109 
110  const Tensor& inp_features = context->input(6);
111 
112  const Tensor& inp_neighbors_importance_sum = context->input(7);
113 
114  const Tensor& inp_neighbors_row_splits = context->input(8);
115 
116  const Tensor& neighbors_index = context->input(9);
117 
118  const Tensor& neighbors_importance = context->input(10);
119 
120  const Tensor& neighbors_row_splits = context->input(11);
121 
122  const Tensor& out_features_gradient = context->input(12);
123 
124  OP_REQUIRES(context, extents.shape().dims() == 2,
125  errors::InvalidArgument("extents must be a rank 2 tensor"));
126  OP_REQUIRES(context,
127  extents.shape().dim_size(0) ==
128  inp_positions.shape().dim_size(0) ||
129  extents.shape().dim_size(0) == 1,
130  errors::InvalidArgument("number of extents must match the "
131  "number of inp_positions or must "
132  "be 1"));
133  OP_REQUIRES(context,
134  extents.shape().dim_size(1) == 3 ||
135  extents.shape().dim_size(1) == 1,
136  errors::InvalidArgument(
137  "number of components for extents must be 3 or 1"));
138 
139  OP_REQUIRES(
140  context,
141  inp_positions.shape().dim_size(0) ==
142  inp_features.shape().dim_size(0),
143  errors::InvalidArgument("first dim of inp_positions does not "
144  "match the first dim of inp_features"));
145 
146  OP_REQUIRES(
147  context,
148  inp_neighbors_importance_sum.shape().dim_size(0) ==
149  inp_positions.shape().dim_size(0) ||
150  inp_neighbors_importance_sum.shape().dim_size(0) == 0,
151  errors::InvalidArgument(
152  "first dim of inp_neighbors_importance_sum does not "
153  "match the first dim of inp_positions",
154  inp_neighbors_importance_sum.shape().dim_size(0), " ",
155  inp_positions.shape().dim_size(0)));
156 
157  OP_REQUIRES(context,
158  out_positions.shape().dim_size(0) ==
159  out_importance.shape().dim_size(0) ||
160  out_importance.shape().dim_size(0) == 0,
161  errors::InvalidArgument("first dim of out_positions does "
162  "not match the first dim of "
163  "out_importance"));
164 
165  OP_REQUIRES(context,
166  neighbors_importance.shape().dim_size(0) ==
167  neighbors_index.shape().dim_size(0) ||
168  neighbors_importance.shape().dim_size(0) == 0,
169  errors::InvalidArgument("first dim of neighbors_importance "
170  "does not match the first dim of "
171  "neighbors_index"));
172 
173  OP_REQUIRES(
174  context,
175  filter.shape().dim_size(3) == inp_features.shape().dim_size(1),
176  errors::InvalidArgument("number of input channels in filter "
177  "and inp_features does not match"));
178 
179  OP_REQUIRES(context,
180  out_features_gradient.shape().dim_size(0) ==
181  out_positions.shape().dim_size(0),
182  errors::InvalidArgument("first dim of out_positions, does "
183  "not match the first dim of "
184  "out_features_gradient"));
185 
186  TensorShape filter_backprop_shape(filter.shape());
187  Tensor* filter_backprop = nullptr;
188  OP_REQUIRES_OK(context,
189  context->allocate_output(0, filter_backprop_shape,
190  &filter_backprop));
191 
192  std::vector<int> filter_dims({
193  int(filter.shape().dim_size(0)),
194  int(filter.shape().dim_size(1)),
195  int(filter.shape().dim_size(2)),
196  int(filter.shape().dim_size(3)),
197  int(filter.shape().dim_size(4)),
198  });
199 
200  bool individual_extents = extents.shape().dim_size(0) ==
201  out_positions.shape().dim_size(0) &&
202  extents.shape().dim_size(0) > 1;
203 
204  bool isotropic_extents = extents.shape().dim_size(1) == 1;
205 
206  bool point_importances = out_importance.shape().dim_size(0) != 0;
207 
208  bool has_neighbors_importances =
209  neighbors_importance.shape().dim_size(0) != 0;
210 
211  Kernel(context, filter, out_positions, out_importance, extents, offset,
212  inp_positions, inp_features, inp_neighbors_importance_sum,
213  inp_neighbors_row_splits, neighbors_index, neighbors_importance,
214  neighbors_row_splits, out_features_gradient, filter_dims,
215  individual_extents, isotropic_extents, point_importances,
216  has_neighbors_importances, *filter_backprop);
217  }
218 
219  virtual void Kernel(tensorflow::OpKernelContext* context,
220  const tensorflow::Tensor& filter,
221  const tensorflow::Tensor& out_positions,
222  const tensorflow::Tensor& out_importance,
223  const tensorflow::Tensor& extents,
224  const tensorflow::Tensor& offset,
225  const tensorflow::Tensor& inp_positions,
226  const tensorflow::Tensor& inp_features,
227  const tensorflow::Tensor& inp_neighbors_importance_sum,
228  const tensorflow::Tensor& inp_neighbors_row_splits,
229  const tensorflow::Tensor& neighbors_index,
230  const tensorflow::Tensor& neighbors_importance,
231  const tensorflow::Tensor& neighbors_row_splits,
232  const tensorflow::Tensor& out_features_gradient,
233  const std::vector<int>& filter_dims,
234  const bool individual_extents,
235  const bool isotropic_extents,
236  const bool point_importances,
237  const bool has_neighbors_importances,
238  tensorflow::Tensor& filter_backprop) = 0;
239 
240 public:
242  bool normalize;
246 };
ImGuiContext * context
Definition: Window.cpp:95
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:36
ContinuousConvTransposeBackpropFilterOpKernel(tensorflow::OpKernelConstruction *construction)
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:38
virtual void Kernel(tensorflow::OpKernelContext *context, const tensorflow::Tensor &filter, const tensorflow::Tensor &out_positions, const tensorflow::Tensor &out_importance, const tensorflow::Tensor &extents, const tensorflow::Tensor &offset, const tensorflow::Tensor &inp_positions, const tensorflow::Tensor &inp_features, const tensorflow::Tensor &inp_neighbors_importance_sum, const tensorflow::Tensor &inp_neighbors_row_splits, const tensorflow::Tensor &neighbors_index, const tensorflow::Tensor &neighbors_importance, const tensorflow::Tensor &neighbors_row_splits, const tensorflow::Tensor &out_features_gradient, const std::vector< int > &filter_dims, const bool individual_extents, const bool isotropic_extents, const bool point_importances, const bool has_neighbors_importances, tensorflow::Tensor &filter_backprop)=0
bool align_corners
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:241
void Compute(tensorflow::OpKernelContext *context) override
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:75
open3d::ml::impl::InterpolationMode interpolation
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:243
open3d::ml::impl::CoordinateMapping coordinate_mapping
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:244
bool normalize
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:242
int max_temp_mem_MB
Definition: ContinuousConvTransposeBackpropFilterOpKernel.h:245
int offset
Definition: FilePCD.cpp:64
const char const char value recording_handle imu_sample recording_handle uint8_t size_t data_size k4a_record_configuration_t config target_format k4a_capture_t capture_handle k4a_imu_sample_t imu_sample playback_handle k4a_logging_message_cb_t void min_level device_handle k4a_imu_sample_t timeout_in_ms capture_handle capture_handle capture_handle image_handle temperature_c int
Definition: K4aPlugin.cpp:493
Definition: ContinuousConv.h:35
InterpolationMode
Definition: ContinuousConvTypes.h:37
@ NEAREST_NEIGHBOR
Definition: VoxelPooling.h:40
CoordinateMapping
Definition: ContinuousConvTypes.h:45