1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
|
/*
* Copyright (c) 2018-2019 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include "HOGDetector.h"
namespace arm_compute
{
namespace test
{
namespace validation
{
namespace reference
{
namespace
{
/** Computes the number of detection windows to iterate over in the feature vector. */
Size2D num_detection_windows(const TensorShape &shape, const Size2D &window_step, const HOGInfo &hog_info)
{
const size_t num_block_strides_width = hog_info.detection_window_size().width / hog_info.block_stride().width;
const size_t num_block_strides_height = hog_info.detection_window_size().height / hog_info.block_stride().height;
return Size2D{ floor_to_multiple(shape.x() - num_block_strides_width, window_step.width) + window_step.width,
floor_to_multiple(shape.y() - num_block_strides_height, window_step.height) + window_step.height };
}
} // namespace
template <typename T>
std::vector<DetectionWindow> hog_detector(const SimpleTensor<T> &src, const std::vector<T> &descriptor, unsigned int max_num_detection_windows,
const HOGInfo &hog_info, const Size2D &detection_window_stride, float threshold, uint16_t idx_class)
{
ARM_COMPUTE_ERROR_ON_MSG((detection_window_stride.width % hog_info.block_stride().width != 0),
"Detection window stride width must be multiple of block stride width");
ARM_COMPUTE_ERROR_ON_MSG((detection_window_stride.height % hog_info.block_stride().height != 0),
"Detection window stride height must be multiple of block stride height");
// Create vector for identifying each detection window
std::vector<DetectionWindow> windows;
// Calculate detection window step
const Size2D window_step(detection_window_stride.width / hog_info.block_stride().width,
detection_window_stride.height / hog_info.block_stride().height);
// Calculate number of detection windows
const Size2D num_windows = num_detection_windows(src.shape(), window_step, hog_info);
// Calculate detection window and row offsets in feature vector
const size_t src_offset_x = window_step.width * hog_info.num_bins() * hog_info.num_cells_per_block().area();
const size_t src_offset_y = window_step.height * hog_info.num_bins() * hog_info.num_cells_per_block().area() * src.shape().x();
const size_t src_offset_row = src.num_channels() * src.shape().x();
// Calculate detection window attributes
const Size2D num_block_positions_per_detection_window = hog_info.num_block_positions_per_image(hog_info.detection_window_size());
const unsigned int num_bins_per_descriptor_x = num_block_positions_per_detection_window.width * src.num_channels();
const unsigned int num_blocks_per_descriptor_y = num_block_positions_per_detection_window.height;
ARM_COMPUTE_ERROR_ON((num_bins_per_descriptor_x * num_blocks_per_descriptor_y + 1) != hog_info.descriptor_size());
size_t win_id = 0;
// Traverse feature vector in detection window steps
for(auto win_y = 0u, offset_y = 0u; win_y < num_windows.height; win_y += window_step.height, offset_y += src_offset_y)
{
for(auto win_x = 0u, offset_x = 0u; win_x < num_windows.width; win_x += window_step.width, offset_x += src_offset_x)
{
// Reset the score
float score = 0.0f;
// Traverse detection window
for(auto y = 0u, offset_row = 0u; y < num_blocks_per_descriptor_y; ++y, offset_row += src_offset_row)
{
const int bin_offset = y * num_bins_per_descriptor_x;
for(auto x = 0u; x < num_bins_per_descriptor_x; ++x)
{
// Compute Linear SVM
const float a = src[x + offset_x + offset_y + offset_row];
const float b = descriptor[x + bin_offset];
score += a * b;
}
}
// Add the bias. The bias is located at the position (descriptor_size() - 1)
score += descriptor[num_bins_per_descriptor_x * num_blocks_per_descriptor_y];
if(score > threshold)
{
DetectionWindow window;
if(win_id++ < max_num_detection_windows)
{
window.x = win_x * hog_info.block_stride().width;
window.y = win_y * hog_info.block_stride().height;
window.width = hog_info.detection_window_size().width;
window.height = hog_info.detection_window_size().height;
window.idx_class = idx_class;
window.score = score;
windows.push_back(window);
}
}
}
}
return windows;
}
template std::vector<DetectionWindow> hog_detector(const SimpleTensor<float> &src, const std::vector<float> &descriptor, unsigned int max_num_detection_windows,
const HOGInfo &hog_info, const Size2D &detection_window_stride, float threshold, uint16_t idx_class);
} // namespace reference
} // namespace validation
} // namespace test
} // namespace arm_compute
|
