30 #include "tensorflow/core/framework/op.h"
31 #include "tensorflow/core/framework/op_kernel.h"
32 #include "tensorflow/core/lib/core/errors.h"
34 template <
class TIndex>
38 tensorflow::OpKernelConstruction* construction)
39 : OpKernel(construction) {
40 using namespace tensorflow;
42 OP_REQUIRES_OK(construction,
44 OP_REQUIRES_OK(construction,
45 construction->GetAttr(
"normalize", &
normalize));
47 std::string interpolation_str;
48 OP_REQUIRES_OK(construction, construction->GetAttr(
"interpolation",
51 if (interpolation_str ==
"linear")
53 else if (interpolation_str ==
"linear_border")
58 std::string mapping_str;
59 OP_REQUIRES_OK(construction, construction->GetAttr(
"coordinate_mapping",
62 if (mapping_str ==
"ball_to_cube_radial")
64 else if (mapping_str ==
"ball_to_cube_volume_preserving")
66 CoordinateMapping::BALL_TO_CUBE_VOLUME_PRESERVING;
70 OP_REQUIRES_OK(construction, construction->GetAttr(
"max_temp_mem_MB",
75 using namespace tensorflow;
76 static_assert(
sizeof(int64) ==
sizeof(int64_t),
77 "int64 type is not compatible");
78 const Tensor& filter =
context->input(0);
80 const Tensor& out_positions =
context->input(1);
82 out_positions.shape().dim_size(0) <=
83 std::numeric_limits<TIndex>::max(),
84 errors::InvalidArgument(
"Too many output points"));
86 const Tensor& extents =
context->input(2);
87 OP_REQUIRES(
context, extents.shape().dims() == 2,
88 errors::InvalidArgument(
"extents must be a rank 2 tensor"));
90 extents.shape().dim_size(0) ==
91 out_positions.shape().dim_size(0) ||
92 extents.shape().dim_size(0) == 1,
93 errors::InvalidArgument(
"number of extents must match the "
94 "number of out_positions or must "
97 extents.shape().dim_size(1) == 3 ||
98 extents.shape().dim_size(1) == 1,
99 errors::InvalidArgument(
100 "number of components for extents must be 3 or 1"));
104 errors::InvalidArgument(
"offset must be a rank 1 tensor"));
106 errors::InvalidArgument(
"offset length must be 3"));
108 const Tensor& inp_positions =
context->input(4);
110 inp_positions.shape().dim_size(0) <=
111 std::numeric_limits<TIndex>::max(),
112 errors::InvalidArgument(
"Too many input points"));
114 const Tensor& inp_features =
context->input(5);
116 const Tensor& inp_importance =
context->input(6);
118 const Tensor& neighbors_index =
context->input(7);
120 const Tensor& neighbors_importance =
context->input(8);
122 const Tensor& neighbors_row_splits =
context->input(9);
124 const Tensor& out_features_gradient =
context->input(10);
128 inp_positions.shape().dim_size(0) ==
129 inp_features.shape().dim_size(0),
130 errors::InvalidArgument(
"first dim of inp_positions does not "
131 "match the first dim of inp_features"));
134 inp_positions.shape().dim_size(0) ==
135 inp_importance.shape().dim_size(0) ||
136 inp_importance.shape().dim_size(0) == 0,
137 errors::InvalidArgument(
"first dim of inp_positions does "
138 "not match the first dim of "
142 neighbors_importance.shape().dim_size(0) ==
143 neighbors_index.shape().dim_size(0) ||
144 neighbors_importance.shape().dim_size(0) == 0,
145 errors::InvalidArgument(
"first dim of neighbors_importance "
146 "does not match the first dim of "
151 filter.shape().dim_size(3) == inp_features.shape().dim_size(1),
152 errors::InvalidArgument(
"number of input channels in filter "
153 "and inp_features does not match"));
156 out_features_gradient.shape().dim_size(0) ==
157 out_positions.shape().dim_size(0),
158 errors::InvalidArgument(
"first dim of out_positions, does "
159 "not match the first dim of "
160 "out_features_gradient"));
162 TensorShape filter_backprop_shape(filter.shape());
163 Tensor* filter_backprop =
nullptr;
165 context->allocate_output(0, filter_backprop_shape,
168 std::vector<int> filter_dims({
169 int(filter.shape().dim_size(0)),
170 int(filter.shape().dim_size(1)),
171 int(filter.shape().dim_size(2)),
172 int(filter.shape().dim_size(3)),
173 int(filter.shape().dim_size(4)),
176 bool individual_extents = extents.shape().dim_size(0) ==
177 out_positions.shape().dim_size(0) &&
178 extents.shape().dim_size(0) > 1;
180 bool isotropic_extents = extents.shape().dim_size(1) == 1;
182 bool point_importances = inp_importance.shape().dim_size(0) != 0;
184 bool has_neighbors_importances =
185 neighbors_importance.shape().dim_size(0) != 0;
188 inp_features, inp_importance, neighbors_index,
189 neighbors_importance, neighbors_row_splits,
190 out_features_gradient, filter_dims, individual_extents,
191 isotropic_extents, point_importances, has_neighbors_importances,
196 const tensorflow::Tensor& filter,
197 const tensorflow::Tensor& out_positions,
198 const tensorflow::Tensor& extents,
199 const tensorflow::Tensor&
offset,
200 const tensorflow::Tensor& inp_positions,
201 const tensorflow::Tensor& inp_features,
202 const tensorflow::Tensor& inp_importance,
203 const tensorflow::Tensor& neighbors_index,
204 const tensorflow::Tensor& neighbors_importance,
205 const tensorflow::Tensor& neighbors_row_splits,
206 const tensorflow::Tensor& out_features_gradient,
207 const std::vector<int>& filter_dims,
208 const bool individual_extents,
209 const bool isotropic_extents,
210 const bool point_importances,
211 const bool has_neighbors_importances,
212 tensorflow::Tensor& filter_backprop) = 0;
ImGuiContext * context
Definition: Window.cpp:95
Definition: ContinuousConvBackpropFilterOpKernel.h:35
bool align_corners
Definition: ContinuousConvBackpropFilterOpKernel.h:215
int max_temp_mem_MB
Definition: ContinuousConvBackpropFilterOpKernel.h:219
open3d::ml::impl::CoordinateMapping coordinate_mapping
Definition: ContinuousConvBackpropFilterOpKernel.h:218
ContinuousConvBackpropFilterOpKernel(tensorflow::OpKernelConstruction *construction)
Definition: ContinuousConvBackpropFilterOpKernel.h:37
bool normalize
Definition: ContinuousConvBackpropFilterOpKernel.h:216
void Compute(tensorflow::OpKernelContext *context) override
Definition: ContinuousConvBackpropFilterOpKernel.h:74
open3d::ml::impl::InterpolationMode interpolation
Definition: ContinuousConvBackpropFilterOpKernel.h:217
virtual void Kernel(tensorflow::OpKernelContext *context, const tensorflow::Tensor &filter, const tensorflow::Tensor &out_positions, const tensorflow::Tensor &extents, const tensorflow::Tensor &offset, const tensorflow::Tensor &inp_positions, const tensorflow::Tensor &inp_features, const tensorflow::Tensor &inp_importance, 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
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