1 /*
2 * Copyright (c) 2018 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 */
24 #include "PriorBoxLayer.h"
25
26 #include "ActivationLayer.h"
27
28 #include "tests/validation/Helpers.h"
29
30 namespace arm_compute
31 {
32 namespace test
33 {
34 namespace validation
35 {
36 namespace reference
37 {
38 template <typename T>
prior_box_layer(const SimpleTensor<T> & src1,const SimpleTensor<T> & src2,const PriorBoxLayerInfo & info,const TensorShape & output_shape)39 SimpleTensor<T> prior_box_layer(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, const PriorBoxLayerInfo &info, const TensorShape &output_shape)
40 {
41 const auto layer_width = static_cast<int>(src1.shape()[0]);
42 const auto layer_height = static_cast<int>(src1.shape()[1]);
43
44 int img_width = info.img_size().x;
45 int img_height = info.img_size().y;
46 if(img_width == 0 || img_height == 0)
47 {
48 img_width = static_cast<int>(src2.shape()[0]);
49 img_height = static_cast<int>(src2.shape()[1]);
50 }
51
52 float step_x = info.steps()[0];
53 float step_y = info.steps()[1];
54 if(step_x == 0.f || step_y == 0.f)
55 {
56 step_x = static_cast<float>(img_width) / layer_width;
57 step_x = static_cast<float>(img_height) / layer_height;
58 }
59
60 // Calculate number of aspect ratios
61 const int num_priors = info.aspect_ratios().size() * info.min_sizes().size() + info.max_sizes().size();
62 const int total_elements = layer_width * layer_height * num_priors * 4;
63
64 SimpleTensor<T> result(output_shape, src1.data_type());
65
66 int idx = 0;
67 for(int y = 0; y < layer_height; ++y)
68 {
69 for(int x = 0; x < layer_width; ++x)
70 {
71 const float center_x = (x + info.offset()) * step_x;
72 const float center_y = (y + info.offset()) * step_y;
73 float box_width;
74 float box_height;
75 for(unsigned int i = 0; i < info.min_sizes().size(); ++i)
76 {
77 const float min_size = info.min_sizes().at(i);
78 box_width = min_size;
79 box_height = min_size;
80 // (xmin, ymin, xmax, ymax)
81 result[idx++] = (center_x - box_width / 2.f) / img_width;
82 result[idx++] = (center_y - box_height / 2.f) / img_height;
83 result[idx++] = (center_x + box_width / 2.f) / img_width;
84 result[idx++] = (center_y + box_height / 2.f) / img_height;
85
86 if(!info.max_sizes().empty())
87 {
88 const float max_size = info.max_sizes().at(i);
89 box_width = sqrt(min_size * max_size);
90 box_height = box_width;
91
92 // (xmin, ymin, xmax, ymax)
93 result[idx++] = (center_x - box_width / 2.f) / img_width;
94 result[idx++] = (center_y - box_height / 2.f) / img_height;
95 result[idx++] = (center_x + box_width / 2.f) / img_width;
96 result[idx++] = (center_y + box_height / 2.f) / img_height;
97 }
98
99 // rest of priors
100 for(auto ar : info.aspect_ratios())
101 {
102 if(fabs(ar - 1.) < 1e-6)
103 {
104 continue;
105 }
106
107 box_width = min_size * sqrt(ar);
108 box_height = min_size / sqrt(ar);
109
110 // (xmin, ymin, xmax, ymax)
111 result[idx++] = (center_x - box_width / 2.f) / img_width;
112 result[idx++] = (center_y - box_height / 2.f) / img_height;
113 result[idx++] = (center_x + box_width / 2.f) / img_width;
114 result[idx++] = (center_y + box_height / 2.f) / img_height;
115 }
116 }
117 }
118 }
119
120 // clip the coordinates
121 if(info.clip())
122 {
123 for(int i = 0; i < total_elements; ++i)
124 {
125 result[i] = std::min<T>(std::max<T>(result[i], 0.f), 1.f);
126 }
127 }
128
129 // set the variance.
130 if(info.variances().size() == 1)
131 {
132 std::fill_n(result.data() + idx, total_elements, info.variances().at(0));
133 }
134 else
135 {
136 for(int h = 0; h < layer_height; ++h)
137 {
138 for(int w = 0; w < layer_width; ++w)
139 {
140 for(int i = 0; i < num_priors; ++i)
141 {
142 for(int j = 0; j < 4; ++j)
143 {
144 result[idx++] = info.variances().at(j);
145 }
146 }
147 }
148 }
149 }
150
151 return result;
152 }
153 template SimpleTensor<float> prior_box_layer(const SimpleTensor<float> &src1, const SimpleTensor<float> &src2, const PriorBoxLayerInfo &info, const TensorShape &output_shape);
154
155 } // namespace reference
156 } // namespace validation
157 } // namespace test
158 } // namespace arm_compute
159