Compute Library
 20.02.1
CLHOGMultiDetection.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2019 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 
27 #include "arm_compute/core/Error.h"
34 
35 using namespace arm_compute;
36 
37 CLHOGMultiDetection::CLHOGMultiDetection(std::shared_ptr<IMemoryManager> memory_manager) // NOLINT
38  : _memory_group(std::move(memory_manager)),
39  _gradient_kernel(),
40  _orient_bin_kernel(),
41  _block_norm_kernel(),
42  _hog_detect_kernel(),
43  _non_maxima_kernel(),
44  _hog_space(),
45  _hog_norm_space(),
46  _detection_windows(),
47  _mag(),
48  _phase(),
49  _non_maxima_suppression(false),
50  _num_orient_bin_kernel(0),
51  _num_block_norm_kernel(0),
52  _num_hog_detect_kernel(0)
53 {
54 }
55 
56 void CLHOGMultiDetection::configure(ICLTensor *input, const ICLMultiHOG *multi_hog, ICLDetectionWindowArray *detection_windows, ICLSize2DArray *detection_window_strides, BorderMode border_mode,
57  uint8_t constant_border_value, float threshold, bool non_maxima_suppression, float min_distance)
58 {
61  ARM_COMPUTE_ERROR_ON(nullptr == detection_windows);
62  ARM_COMPUTE_ERROR_ON(detection_window_strides->num_values() != multi_hog->num_models());
63 
64  const size_t width = input->info()->dimension(Window::DimX);
65  const size_t height = input->info()->dimension(Window::DimY);
66  const TensorShape &shape_img = input->info()->tensor_shape();
67  const size_t num_models = multi_hog->num_models();
68  PhaseType phase_type = multi_hog->model(0)->info()->phase_type();
69 
70  size_t prev_num_bins = multi_hog->model(0)->info()->num_bins();
71  Size2D prev_cell_size = multi_hog->model(0)->info()->cell_size();
72  Size2D prev_block_size = multi_hog->model(0)->info()->block_size();
73  Size2D prev_block_stride = multi_hog->model(0)->info()->block_stride();
74 
75  /* Check if CLHOGOrientationBinningKernel and CLHOGBlockNormalizationKernel kernels can be skipped for a specific HOG data-object
76  *
77  * 1) CLHOGOrientationBinningKernel and CLHOGBlockNormalizationKernel are skipped if the cell size and the number of bins don't change.
78  * Since "multi_hog" is sorted,it is enough to check the HOG descriptors at level "ith" and level "(i-1)th
79  * 2) CLHOGBlockNormalizationKernel is skipped if the cell size, the number of bins and block size do not change.
80  * Since "multi_hog" is sorted,it is enough to check the HOG descriptors at level "ith" and level "(i-1)th
81  *
82  * @note Since the orientation binning and block normalization kernels can be skipped, we need to keep track of the input to process for each kernel
83  * with "input_orient_bin", "input_hog_detect" and "input_block_norm"
84  */
85  std::vector<size_t> input_orient_bin;
86  std::vector<size_t> input_hog_detect;
87  std::vector<std::pair<size_t, size_t>> input_block_norm;
88 
89  input_orient_bin.push_back(0);
90  input_hog_detect.push_back(0);
91  input_block_norm.emplace_back(0, 0);
92 
93  for(size_t i = 1; i < num_models; ++i)
94  {
95  size_t cur_num_bins = multi_hog->model(i)->info()->num_bins();
96  Size2D cur_cell_size = multi_hog->model(i)->info()->cell_size();
97  Size2D cur_block_size = multi_hog->model(i)->info()->block_size();
98  Size2D cur_block_stride = multi_hog->model(i)->info()->block_stride();
99 
100  if((cur_num_bins != prev_num_bins) || (cur_cell_size.width != prev_cell_size.width) || (cur_cell_size.height != prev_cell_size.height))
101  {
102  prev_num_bins = cur_num_bins;
103  prev_cell_size = cur_cell_size;
104  prev_block_size = cur_block_size;
105  prev_block_stride = cur_block_stride;
106 
107  // Compute orientation binning and block normalization kernels. Update input to process
108  input_orient_bin.push_back(i);
109  input_block_norm.emplace_back(i, input_orient_bin.size() - 1);
110  }
111  else if((cur_block_size.width != prev_block_size.width) || (cur_block_size.height != prev_block_size.height) || (cur_block_stride.width != prev_block_stride.width)
112  || (cur_block_stride.height != prev_block_stride.height))
113  {
114  prev_block_size = cur_block_size;
115  prev_block_stride = cur_block_stride;
116 
117  // Compute block normalization kernel. Update input to process
118  input_block_norm.emplace_back(i, input_orient_bin.size() - 1);
119  }
120 
121  // Update input to process for hog detector kernel
122  input_hog_detect.push_back(input_block_norm.size() - 1);
123  }
124 
125  _detection_windows = detection_windows;
126  _non_maxima_suppression = non_maxima_suppression;
127  _num_orient_bin_kernel = input_orient_bin.size(); // Number of CLHOGOrientationBinningKernel kernels to compute
128  _num_block_norm_kernel = input_block_norm.size(); // Number of CLHOGBlockNormalizationKernel kernels to compute
129  _num_hog_detect_kernel = input_hog_detect.size(); // Number of CLHOGDetector functions to compute
130 
131  _orient_bin_kernel.resize(_num_orient_bin_kernel);
132  _block_norm_kernel.resize(_num_block_norm_kernel);
133  _hog_detect_kernel.resize(_num_hog_detect_kernel);
134  _hog_space.resize(_num_orient_bin_kernel);
135  _hog_norm_space.resize(_num_block_norm_kernel);
136 
137  // Allocate tensors for magnitude and phase
138  TensorInfo info_mag(shape_img, Format::S16);
139  _mag.allocator()->init(info_mag);
140 
141  TensorInfo info_phase(shape_img, Format::U8);
142  _phase.allocator()->init(info_phase);
143 
144  // Manage intermediate buffers
145  _memory_group.manage(&_mag);
146  _memory_group.manage(&_phase);
147 
148  // Initialise gradient kernel
149  _gradient_kernel.configure(input, &_mag, &_phase, phase_type, border_mode, constant_border_value);
150 
151  // Configure NETensor for the HOG space and orientation binning kernel
152  for(size_t i = 0; i < _num_orient_bin_kernel; ++i)
153  {
154  const size_t idx_multi_hog = input_orient_bin[i];
155 
156  // Get the corresponding cell size and number of bins
157  const Size2D &cell = multi_hog->model(idx_multi_hog)->info()->cell_size();
158  const size_t num_bins = multi_hog->model(idx_multi_hog)->info()->num_bins();
159 
160  // Calculate number of cells along the x and y directions for the hog_space
161  const size_t num_cells_x = width / cell.width;
162  const size_t num_cells_y = height / cell.height;
163 
164  // TensorShape of hog space
165  TensorShape shape_hog_space = input->info()->tensor_shape();
166  shape_hog_space.set(Window::DimX, num_cells_x);
167  shape_hog_space.set(Window::DimY, num_cells_y);
168 
169  // Allocate HOG space
170  TensorInfo info_space(shape_hog_space, num_bins, DataType::F32);
171  _hog_space[i].allocator()->init(info_space);
172 
173  // Manage intermediate buffers
174  _memory_group.manage(&_hog_space[i]);
175 
176  // Initialise orientation binning kernel
177  _orient_bin_kernel[i].configure(&_mag, &_phase, &_hog_space[i], multi_hog->model(idx_multi_hog)->info());
178  }
179 
180  // Allocate intermediate tensors
181  _mag.allocator()->allocate();
182  _phase.allocator()->allocate();
183 
184  // Configure CLTensor for the normalized HOG space and block normalization kernel
185  for(size_t i = 0; i < _num_block_norm_kernel; ++i)
186  {
187  const size_t idx_multi_hog = input_block_norm[i].first;
188  const size_t idx_orient_bin = input_block_norm[i].second;
189 
190  // Allocate normalized HOG space
191  TensorInfo tensor_info(*(multi_hog->model(idx_multi_hog)->info()), width, height);
192  _hog_norm_space[i].allocator()->init(tensor_info);
193 
194  // Manage intermediate buffers
195  _memory_group.manage(&_hog_norm_space[i]);
196 
197  // Initialize block normalization kernel
198  _block_norm_kernel[i].configure(&_hog_space[idx_orient_bin], &_hog_norm_space[i], multi_hog->model(idx_multi_hog)->info());
199  }
200 
201  // Allocate intermediate tensors
202  for(size_t i = 0; i < _num_orient_bin_kernel; ++i)
203  {
204  _hog_space[i].allocator()->allocate();
205  }
206 
207  detection_window_strides->map(CLScheduler::get().queue(), true);
208 
209  // Configure HOG detector kernel
210  for(size_t i = 0; i < _num_hog_detect_kernel; ++i)
211  {
212  const size_t idx_block_norm = input_hog_detect[i];
213 
214  _hog_detect_kernel[i].configure(&_hog_norm_space[idx_block_norm], multi_hog->cl_model(i), detection_windows, detection_window_strides->at(i), threshold, i);
215  }
216 
217  detection_window_strides->unmap(CLScheduler::get().queue());
218 
219  // Configure non maxima suppression kernel
220  _non_maxima_kernel.configure(_detection_windows, min_distance);
221 
222  // Allocate intermediate tensors
223  for(size_t i = 0; i < _num_block_norm_kernel; ++i)
224  {
225  _hog_norm_space[i].allocator()->allocate();
226  }
227 }
228 
230 {
231  ARM_COMPUTE_ERROR_ON_MSG(_detection_windows == nullptr, "Unconfigured function");
232 
233  MemoryGroupResourceScope scope_mg(_memory_group);
234 
235  // Reset detection window
236  _detection_windows->clear();
237 
238  // Run gradient
239  _gradient_kernel.run();
240 
241  // Run orientation binning kernel
242  for(size_t i = 0; i < _num_orient_bin_kernel; ++i)
243  {
244  CLScheduler::get().enqueue(_orient_bin_kernel[i], false);
245  }
246 
247  // Run block normalization kernel
248  for(size_t i = 0; i < _num_block_norm_kernel; ++i)
249  {
250  CLScheduler::get().enqueue(_block_norm_kernel[i], false);
251  }
252 
253  // Run HOG detector kernel
254  for(size_t i = 0; i < _num_hog_detect_kernel; ++i)
255  {
256  _hog_detect_kernel[i].run();
257  }
258 
259  // Run non-maxima suppression kernel if enabled
260  if(_non_maxima_suppression)
261  {
262  // Map detection windows array before computing non maxima suppression
263  _detection_windows->map(CLScheduler::get().queue(), true);
264  Scheduler::get().schedule(&_non_maxima_kernel, Window::DimY);
265  _detection_windows->unmap(CLScheduler::get().queue());
266  }
267 }
BorderMode
Methods available to handle borders.
Definition: Types.h:261
void configure(ICLTensor *input, const ICLMultiHOG *multi_hog, ICLDetectionWindowArray *detection_windows, ICLSize2DArray *detection_window_strides, BorderMode border_mode, uint8_t constant_border_value=0, float threshold=0.0f, bool non_maxima_suppression=false, float min_distance=1.0f)
Initialise the function's source, destination, detection window strides, border mode,...
size_t num_bins() const
The number of histogram bins for each cell.
Definition: HOGInfo.cpp:111
void clear()
Clear all the points from the array.
Definition: IArray.h:91
Shape of a tensor.
Definition: TensorShape.h:39
const Size2D & cell_size() const
The cell size in pixels.
Definition: HOGInfo.cpp:91
static CLScheduler & get()
Access the scheduler singleton.
Definition: CLScheduler.cpp:99
PhaseType
Phase calculation type.
Definition: Types.h:406
void configure(ICLTensor *input, ICLTensor *output_magnitude, ICLTensor *output_phase, PhaseType phase_type, BorderMode border_mode, uint8_t constant_border_value=0)
Initialise the function's source, destinations, phase type and border mode.
1 channel, 1 U8 per channel
1 channel, 1 F32 per channel
void map(cl::CommandQueue &q, bool blocking=true)
Enqueue a map operation of the allocated buffer on the given queue.
Definition: ICLArray.h:72
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:466
CLHOGMultiDetection(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Default constructor.
CLTensorAllocator * allocator()
Return a pointer to the tensor's allocator.
Definition: CLTensor.cpp:61
PhaseType phase_type() const
The type of PhaseType.
Definition: HOGInfo.cpp:126
virtual ICLHOG * cl_model(size_t index)=0
Return a pointer to the requested OpenCL HOG model.
const Size2D & block_stride() const
The block stride in pixels.
Definition: HOGInfo.cpp:106
void init(const TensorInfo &input, size_t alignment=0)
Initialize a tensor based on the passed TensorInfo.
Copyright (c) 2017-2020 ARM Limited.
size_t height
Height of the image region or rectangle.
Definition: Size2D.h:93
#define ARM_COMPUTE_ERROR_ON_INVALID_MULTI_HOG(m)
Definition: Validate.h:925
virtual size_t num_models() const =0
The number of HOG models stored.
void manage(IMemoryManageable *obj) override
Sets a object to be managed by the given memory group.
Definition: MemoryGroup.h:79
IHOG * model(size_t index) override
Return a pointer to the requested HOG model.
Definition: ICLMultiHOG.cpp:30
Interface for OpenCL Array.
Definition: ICLArray.h:35
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
size_t num_values() const
Number of values currently stored in the array.
Definition: IArray.h:68
Interface for storing multiple HOG data-objects.
Definition: ICLMultiHOG.h:33
#define ARM_COMPUTE_ERROR_ON_MSG(cond, msg)
Definition: Error.h:456
const Size2D & block_size() const
The block size in pixels.
Definition: HOGInfo.cpp:96
void configure(IDetectionWindowArray *input_output, float min_distance)
Initialise the kernel's input, output and the euclidean minimum distance.
1 channel, 1 S16 per channel
void enqueue(ICLKernel &kernel, bool flush=true)
Schedule the execution of the passed kernel if possible.
#define ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:790
void unmap(cl::CommandQueue &q)
Enqueue an unmap operation of the allocated and mapped buffer on the given queue.
Definition: ICLArray.h:83
void allocate() override
Allocate size specified by TensorInfo of OpenCL memory.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
Memory group resources scope handling class.
Definition: IMemoryGroup.h:82
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
virtual void schedule(ICPPKernel *kernel, const Hints &hints)=0
Runs the kernel in the same thread as the caller synchronously.
size_t width
Width of the image region or rectangle.
Definition: Size2D.h:92
SimpleTensor< T > non_maxima_suppression(const SimpleTensor< T > &src, BorderMode border_mode, T constant_border_value)
Class for specifying the size of an image or rectangle.
Definition: Size2D.h:34
TensorShape & set(size_t dimension, size_t value, bool apply_dim_correction=true)
Accessor to set the value of one of the dimensions.
Definition: TensorShape.h:78
Store the tensor's metadata.
Definition: TensorInfo.h:45
virtual const HOGInfo * info() const =0
Interface to be implemented by the child class to return the HOG's metadata.
void run() override
Run the kernels contained in the function.
SimpleTensor< T > threshold(const SimpleTensor< T > &src, T threshold, T false_value, T true_value, ThresholdType type, T upper)
Definition: Threshold.cpp:35
void run() override
Run the kernels contained in the function.
virtual T & at(size_t index) const
Reference to the element of the array located at the given index.
Definition: IArray.h:117
static IScheduler & get()
Access the scheduler singleton.
Definition: Scheduler.cpp:95