Compute Library
 21.05
CPPNonMaximumSuppressionKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019-2020 Arm Limited.
3  *
4  * SPDX-License-Identifier: MIT
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to
8  * deal in the Software without restriction, including without limitation the
9  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10  * sell copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in all
14  * copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
25 
28 
31 
32 #include <algorithm>
33 
34 namespace arm_compute
35 {
36 namespace
37 {
38 Status validate_arguments(const ITensorInfo *bboxes, const ITensorInfo *scores, const ITensorInfo *output_indices, unsigned int max_output_size,
39  const float score_threshold, const float iou_threshold)
40 {
41  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(bboxes, scores, output_indices);
44  ARM_COMPUTE_RETURN_ERROR_ON_MSG(bboxes->num_dimensions() > 2, "The bboxes tensor must be a 2-D float tensor of shape [4, num_boxes].");
45  ARM_COMPUTE_RETURN_ERROR_ON_MSG(scores->num_dimensions() > 1, "The scores tensor must be a 1-D float tensor of shape [num_boxes].");
46  ARM_COMPUTE_RETURN_ERROR_ON_MSG(output_indices->num_dimensions() > 1, "The indices must be 1-D integer tensor of shape [M], where max_output_size <= M");
48  ARM_COMPUTE_RETURN_ERROR_ON_MSG(output_indices->dimension(0) == 0, "Indices tensor must be bigger than 0");
49  ARM_COMPUTE_RETURN_ERROR_ON_MSG(max_output_size == 0, "Max size cannot be 0");
50  ARM_COMPUTE_RETURN_ERROR_ON_MSG(iou_threshold < 0.f || iou_threshold > 1.f, "IOU threshold must be in [0,1]");
51  ARM_COMPUTE_RETURN_ERROR_ON_MSG(score_threshold < 0.f || score_threshold > 1.f, "Score threshold must be in [0,1]");
52 
53  return Status{};
54 }
55 } // namespace
56 
58  : _input_bboxes(nullptr), _input_scores(nullptr), _output_indices(nullptr), _max_output_size(0), _score_threshold(0.f), _iou_threshold(0.f), _num_boxes(0)
59 {
60 }
61 
62 void CPPNonMaximumSuppressionKernel::configure(const ITensor *input_bboxes, const ITensor *input_scores, ITensor *output_indices,
63  unsigned int max_output_size, const float score_threshold, const float iou_threshold)
64 {
65  ARM_COMPUTE_ERROR_ON_NULLPTR(input_bboxes, input_scores, output_indices);
66  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input_bboxes->info(), input_scores->info(), output_indices->info(), max_output_size, score_threshold, iou_threshold));
67 
68  auto_init_if_empty(*output_indices->info(), TensorShape(max_output_size), 1, DataType::U8, QuantizationInfo());
69 
70  _input_bboxes = input_bboxes;
71  _input_scores = input_scores;
72  _output_indices = output_indices;
73  _score_threshold = score_threshold;
74  _iou_threshold = iou_threshold;
75  _max_output_size = max_output_size;
76  _num_boxes = input_scores->info()->dimension(0);
77 
78  // Configure kernel window
79  Window win = calculate_max_window(*output_indices->info(), Steps());
80 
81  // The CPPNonMaximumSuppressionKernel doesn't need padding so update_window_and_padding() can be skipped
82  ICPPKernel::configure(win);
83 }
84 
85 Status CPPNonMaximumSuppressionKernel::validate(const ITensorInfo *bboxes, const ITensorInfo *scores, const ITensorInfo *output_indices,
86  unsigned int max_output_size, const float score_threshold, const float iou_threshold)
87 {
88  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(bboxes, scores, output_indices, max_output_size, score_threshold, iou_threshold));
89  return Status{};
90 }
91 
93 {
98 
99  // Auxiliary tensors
100  std::vector<int> indices_above_thd;
101  std::vector<float> scores_above_thd;
102  for(unsigned int i = 0; i < _num_boxes; ++i)
103  {
104  const float score_i = *(reinterpret_cast<float *>(_input_scores->ptr_to_element(Coordinates(i))));
105  if(score_i >= _score_threshold)
106  {
107  scores_above_thd.emplace_back(score_i);
108  indices_above_thd.emplace_back(i);
109  }
110  }
111 
112  // Sort selected indices based on scores
113  const unsigned int num_above_thd = indices_above_thd.size();
114  std::vector<unsigned int> sorted_indices;
115  sorted_indices.resize(num_above_thd);
116  std::iota(sorted_indices.data(), sorted_indices.data() + num_above_thd, 0);
117  std::sort(std::begin(sorted_indices),
118  std::end(sorted_indices),
119  [&](unsigned int first, unsigned int second)
120  {
121  return scores_above_thd[first] > scores_above_thd[second];
122  });
123 
124  // Number of output is the minimum between max_detection and the scores above the threshold
125  const unsigned int num_output = std::min(_max_output_size, num_above_thd);
126  unsigned int output_idx = 0;
127  std::vector<bool> visited(num_above_thd, false);
128 
129  // Keep only boxes with small IoU
130  for(unsigned int i = 0; i < num_above_thd; ++i)
131  {
132  // Check if the output is full
133  if(output_idx >= num_output)
134  {
135  break;
136  }
137 
138  // Check if it was already visited, if not add it to the output and update the indices counter
139  if(!visited[sorted_indices[i]])
140  {
141  *(reinterpret_cast<int *>(_output_indices->ptr_to_element(Coordinates(output_idx)))) = indices_above_thd[sorted_indices[i]];
142  visited[sorted_indices[i]] = true;
143  ++output_idx;
144  }
145  else
146  {
147  continue;
148  }
149 
150  // Once added one element at the output check if the next ones overlap and can be skipped
151  for(unsigned int j = i + 1; j < num_above_thd; ++j)
152  {
153  if(!visited[sorted_indices[j]])
154  {
155  // Calculate IoU
156  const unsigned int i_index = indices_above_thd[sorted_indices[i]];
157  const unsigned int j_index = indices_above_thd[sorted_indices[j]];
158  // Box-corner format: xmin, ymin, xmax, ymax
159  const auto box_i_xmin = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(0, i_index))));
160  const auto box_i_ymin = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(1, i_index))));
161  const auto box_i_xmax = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(2, i_index))));
162  const auto box_i_ymax = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(3, i_index))));
163 
164  const auto box_j_xmin = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(0, j_index))));
165  const auto box_j_ymin = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(1, j_index))));
166  const auto box_j_xmax = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(2, j_index))));
167  const auto box_j_ymax = *(reinterpret_cast<float *>(_input_bboxes->ptr_to_element(Coordinates(3, j_index))));
168 
169  const float area_i = (box_i_xmax - box_i_xmin) * (box_i_ymax - box_i_ymin);
170  const float area_j = (box_j_xmax - box_j_xmin) * (box_j_ymax - box_j_ymin);
171  float overlap;
172  if(area_i <= 0 || area_j <= 0)
173  {
174  overlap = 0.0f;
175  }
176  else
177  {
178  const auto y_min_intersection = std::max<float>(box_i_ymin, box_j_ymin);
179  const auto x_min_intersection = std::max<float>(box_i_xmin, box_j_xmin);
180  const auto y_max_intersection = std::min<float>(box_i_ymax, box_j_ymax);
181  const auto x_max_intersection = std::min<float>(box_i_xmax, box_j_xmax);
182  const auto area_intersection = std::max<float>(y_max_intersection - y_min_intersection, 0.0f) * std::max<float>(x_max_intersection - x_min_intersection, 0.0f);
183  overlap = area_intersection / (area_i + area_j - area_intersection);
184  }
185 
186  if(overlap > _iou_threshold)
187  {
188  visited[sorted_indices[j]] = true;
189  }
190  }
191  }
192  }
193  // The output could be full but not the output indices tensor
194  // Instead return values not valid we put -1
195  for(; output_idx < _max_output_size; ++output_idx)
196  {
197  *(reinterpret_cast<int *>(_output_indices->ptr_to_element(Coordinates(output_idx)))) = -1;
198  }
199 }
200 } // namespace arm_compute
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
uint8_t * ptr_to_element(const Coordinates &id) const
Return a pointer to the element at the passed coordinates.
Definition: ITensor.h:63
Shape of a tensor.
Definition: TensorShape.h:39
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
void run(const Window &window, const ThreadInfo &info) override
Execute the kernel on the passed window.
1 channel, 1 U8 per channel
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
static Status validate(const ITensorInfo *input_bboxes, const ITensorInfo *input_scores, const ITensorInfo *output_indices, unsigned int max_output_size, const float score_threshold, const float iou_threshold)
Static function to check if given arguments will lead to a valid configuration of CPPNonMaximumSuppre...
1 channel, 1 F32 per channel
Store the tensor's metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Status class.
Definition: Error.h:52
Interface for CPU tensor.
Definition: ITensor.h:36
Copyright (c) 2017-2021 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
1 channel, 1 S32 per channel
Quantization information.
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Coordinates of an item.
Definition: Coordinates.h:37
bool auto_init_if_empty(ITensorInfo &info, const TensorShape &shape, int num_channels, DataType data_type, QuantizationInfo quantization_info=QuantizationInfo())
Auto initialize the tensor info (shape, number of channels and data type) if the current assignment i...
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
void end(TokenStream &in, bool &valid)
Definition: MLGOParser.cpp:290
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
Information about executing thread and CPU.
Definition: CPPTypes.h:252
void configure(const ITensor *input_bboxes, const ITensor *input_scores, ITensor *output_indices, unsigned int max_output_size, const float score_threshold, const float iou_threshold)
Configure the kernel to perform non maximal suppression.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:788
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201