40 #ifndef PCL_FILTERS_IMPL_FAST_BILATERAL_OMP_HPP_
41 #define PCL_FILTERS_IMPL_FAST_BILATERAL_OMP_HPP_
43 #include <pcl/filters/fast_bilateral_omp.h>
44 #include <pcl/common/io.h>
48 template <
typename Po
intT>
void
53 threads_ = omp_get_num_procs();
58 threads_ = nr_threads;
62 template <
typename Po
intT>
void
65 if (!input_->isOrganized ())
67 PCL_ERROR (
"[pcl::FastBilateralFilterOMP] Input cloud needs to be organized.\n");
72 float base_max = -std::numeric_limits<float>::max (),
73 base_min = std::numeric_limits<float>::max ();
74 bool found_finite =
false;
75 for (std::size_t x = 0; x < output.
width; ++x)
77 for (std::size_t y = 0; y < output.
height; ++y)
79 if (std::isfinite (output (x, y).z))
81 if (base_max < output (x, y).z)
82 base_max = output (x, y).z;
83 if (base_min > output (x, y).z)
84 base_min = output (x, y).z;
91 PCL_WARN (
"[pcl::FastBilateralFilterOMP] Given an empty cloud. Doing nothing.\n");
95 #pragma omp parallel for num_threads (threads_)
97 for (
long int i = 0; i < static_cast<long int> (output.
size ()); ++i)
98 if (!std::isfinite (output.
at(i).z))
99 output.
at(i).z = base_max;
101 const float base_delta = base_max - base_min;
103 const std::size_t padding_xy = 2;
104 const std::size_t padding_z = 2;
106 const std::size_t small_width = static_cast<std::size_t> (static_cast<float> (input_->width - 1) / sigma_s_) + 1 + 2 * padding_xy;
107 const std::size_t small_height = static_cast<std::size_t> (static_cast<float> (input_->height - 1) / sigma_s_) + 1 + 2 * padding_xy;
108 const std::size_t small_depth = static_cast<std::size_t> (base_delta / sigma_r_) + 1 + 2 * padding_z;
110 Array3D data (small_width, small_height, small_depth);
112 #pragma omp parallel for num_threads (threads_)
114 for (
long int i = 0; i < static_cast<long int> (small_width * small_height); ++i)
116 std::size_t small_x = static_cast<std::size_t> (i % small_width);
117 std::size_t small_y = static_cast<std::size_t> (i / small_width);
118 std::size_t start_x = static_cast<std::size_t>(
119 std::max ((static_cast<float> (small_x) - static_cast<float> (padding_xy) - 0.5f) * sigma_s_ + 1, 0.f));
120 std::size_t end_x = static_cast<std::size_t>(
121 std::max ((static_cast<float> (small_x) - static_cast<float> (padding_xy) + 0.5f) * sigma_s_ + 1, 0.f));
122 std::size_t start_y = static_cast<std::size_t>(
123 std::max ((static_cast<float> (small_y) - static_cast<float> (padding_xy) - 0.5f) * sigma_s_ + 1, 0.f));
124 std::size_t end_y = static_cast<std::size_t>(
125 std::max ((static_cast<float> (small_y) - static_cast<float> (padding_xy) + 0.5f) * sigma_s_ + 1, 0.f));
126 for (std::size_t x = start_x; x < end_x && x < input_->width; ++x)
128 for (std::size_t y = start_y; y < end_y && y < input_->height; ++y)
130 const float z = output (x,y).z - base_min;
131 const std::size_t small_z = static_cast<std::size_t> (static_cast<float> (z) / sigma_r_ + 0.5f) + padding_z;
132 Eigen::Vector2f& d = data (small_x, small_y, small_z);
133 d[0] += output (x,y).z;
139 std::vector<long int> offset (3);
140 offset[0] = &(data (1,0,0)) - &(data (0,0,0));
141 offset[1] = &(data (0,1,0)) - &(data (0,0,0));
142 offset[2] = &(data (0,0,1)) - &(data (0,0,0));
144 Array3D buffer (small_width, small_height, small_depth);
146 for (std::size_t dim = 0; dim < 3; ++dim)
148 for (std::size_t n_iter = 0; n_iter < 2; ++n_iter)
150 Array3D* current_buffer = (n_iter % 2 == 1 ? &buffer : &data);
151 Array3D* current_data =(n_iter % 2 == 1 ? &data : &buffer);
153 #pragma omp parallel for num_threads (threads_)
155 for(
long int i = 0; i < static_cast<long int> ((small_width - 2)*(small_height - 2)); ++i)
157 std::size_t x = static_cast<std::size_t> (i % (small_width - 2) + 1);
158 std::size_t y = static_cast<std::size_t> (i / (small_width - 2) + 1);
159 const long int off = offset[dim];
160 Eigen::Vector2f* d_ptr = &(current_data->operator() (x,y,1));
161 Eigen::Vector2f* b_ptr = &(current_buffer->operator() (x,y,1));
163 for(std::size_t z = 1; z < small_depth - 1; ++z, ++d_ptr, ++b_ptr)
164 *d_ptr = (*(b_ptr - off) + *(b_ptr + off) + 2.0 * (*b_ptr)) / 4.0;
174 for (std::vector<Eigen::Vector2f, Eigen::aligned_allocator<Eigen::Vector2f> >::iterator d = data.begin (); d != data.end (); ++d)
175 *d /= ((*d)[0] != 0) ? (*d)[1] : 1;
178 #pragma omp parallel for num_threads (threads_)
180 for (
long int i = 0; i < static_cast<long int> (input_->size ()); ++i)
182 std::size_t x = static_cast<std::size_t> (i % input_->width);
183 std::size_t y = static_cast<std::size_t> (i / input_->width);
184 const float z = output (x,y).z - base_min;
185 const Eigen::Vector2f D = data.trilinear_interpolation (static_cast<float> (x) / sigma_s_ + padding_xy,
186 static_cast<float> (y) / sigma_s_ + padding_xy,
187 z / sigma_r_ + padding_z);
188 output(x,y).z = D[0];
194 #pragma omp parallel for num_threads (threads_)
196 for (
long i = 0; i < static_cast<long int> (input_->size ()); ++i)
198 std::size_t x = static_cast<std::size_t> (i % input_->width);
199 std::size_t y = static_cast<std::size_t> (i / input_->width);
200 const float z = output (x,y).z - base_min;
201 const Eigen::Vector2f D = data.trilinear_interpolation (static_cast<float> (x) / sigma_s_ + padding_xy,
202 static_cast<float> (y) / sigma_s_ + padding_xy,
203 z / sigma_r_ + padding_z);
204 output (x,y).z = D[0] / D[1];