• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2018-2021 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 #ifndef ARM_COMPUTE_TEST_BOUNDINGBOXTRANSFORM_FIXTURE
25 #define ARM_COMPUTE_TEST_BOUNDINGBOXTRANSFORM_FIXTURE
26 
27 #include "arm_compute/core/TensorShape.h"
28 #include "arm_compute/core/Types.h"
29 #include "tests/AssetsLibrary.h"
30 #include "tests/Globals.h"
31 #include "tests/IAccessor.h"
32 #include "tests/framework/Asserts.h"
33 #include "tests/framework/Fixture.h"
34 #include "tests/validation/Helpers.h"
35 #include "tests/validation/reference/BoundingBoxTransform.h"
36 
37 namespace arm_compute
38 {
39 namespace test
40 {
41 namespace validation
42 {
43 namespace
44 {
generate_deltas(std::vector<float> & boxes,const TensorShape & image_shape,size_t num_boxes,size_t num_classes,std::mt19937 & gen)45 std::vector<float> generate_deltas(std::vector<float> &boxes, const TensorShape &image_shape, size_t num_boxes, size_t num_classes, std::mt19937 &gen)
46 {
47     std::vector<float> deltas(num_boxes * 4 * num_classes);
48 
49     std::uniform_int_distribution<> dist_x1(0, image_shape[0] - 1);
50     std::uniform_int_distribution<> dist_y1(0, image_shape[1] - 1);
51     std::uniform_int_distribution<> dist_w(1, image_shape[0]);
52     std::uniform_int_distribution<> dist_h(1, image_shape[1]);
53 
54     for(size_t i = 0; i < num_boxes; ++i)
55     {
56         const float ex_width  = boxes[4 * i + 2] - boxes[4 * i] + 1.f;
57         const float ex_height = boxes[4 * i + 3] - boxes[4 * i + 1] + 1.f;
58         const float ex_ctr_x  = boxes[4 * i] + 0.5f * ex_width;
59         const float ex_ctr_y  = boxes[4 * i + 1] + 0.5f * ex_height;
60 
61         for(size_t j = 0; j < num_classes; ++j)
62         {
63             const float x1     = dist_x1(gen);
64             const float y1     = dist_y1(gen);
65             const float width  = dist_w(gen);
66             const float height = dist_h(gen);
67             const float ctr_x  = x1 + 0.5f * width;
68             const float ctr_y  = y1 + 0.5f * height;
69 
70             deltas[4 * num_classes * i + 4 * j]     = (ctr_x - ex_ctr_x) / ex_width;
71             deltas[4 * num_classes * i + 4 * j + 1] = (ctr_y - ex_ctr_y) / ex_height;
72             deltas[4 * num_classes * i + 4 * j + 2] = log(width / ex_width);
73             deltas[4 * num_classes * i + 4 * j + 3] = log(height / ex_height);
74         }
75     }
76     return deltas;
77 }
78 
generate_boxes(const TensorShape & image_shape,size_t num_boxes,std::mt19937 & gen)79 std::vector<float> generate_boxes(const TensorShape &image_shape, size_t num_boxes, std::mt19937 &gen)
80 {
81     std::vector<float> boxes(num_boxes * 4);
82 
83     std::uniform_int_distribution<> dist_x1(0, image_shape[0] - 1);
84     std::uniform_int_distribution<> dist_y1(0, image_shape[1] - 1);
85     std::uniform_int_distribution<> dist_w(1, image_shape[0]);
86     std::uniform_int_distribution<> dist_h(1, image_shape[1]);
87 
88     for(size_t i = 0; i < num_boxes; ++i)
89     {
90         boxes[4 * i]     = dist_x1(gen);
91         boxes[4 * i + 1] = dist_y1(gen);
92         boxes[4 * i + 2] = boxes[4 * i] + dist_w(gen) - 1;
93         boxes[4 * i + 3] = boxes[4 * i + 1] + dist_h(gen) - 1;
94     }
95     return boxes;
96 }
97 } // namespace
98 
99 template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
100 class BoundingBoxTransformGenericFixture : public framework::Fixture
101 {
102 public:
103     using TDeltas = typename std::conditional<std::is_same<typename std::decay<T>::type, uint16_t>::value, uint8_t, T>::type;
104 
105     template <typename...>
setup(TensorShape deltas_shape,const BoundingBoxTransformInfo & info,DataType data_type,QuantizationInfo deltas_qinfo)106     void setup(TensorShape deltas_shape, const BoundingBoxTransformInfo &info, DataType data_type, QuantizationInfo deltas_qinfo)
107     {
108         const bool is_qasymm16 = data_type == DataType::QASYMM16;
109         _data_type_deltas      = (is_qasymm16) ? DataType::QASYMM8 : data_type;
110         _boxes_qinfo           = (is_qasymm16) ? QuantizationInfo(.125f, 0) : QuantizationInfo();
111 
112         std::mt19937 gen_target(library->seed());
113         _target = compute_target(deltas_shape, data_type, info, gen_target, deltas_qinfo);
114 
115         std::mt19937 gen_reference(library->seed());
116         _reference = compute_reference(deltas_shape, data_type, info, gen_reference, deltas_qinfo);
117     }
118 
119 protected:
120     template <typename data_type, typename U>
fill(U && tensor,std::vector<float> values)121     void fill(U &&tensor, std::vector<float> values)
122     {
123         data_type *data_ptr = reinterpret_cast<data_type *>(tensor.data());
124         switch(tensor.data_type())
125         {
126             case DataType::QASYMM8:
127                 for(size_t i = 0; i < values.size(); ++i)
128                 {
129                     data_ptr[i] = quantize_qasymm8(values[i], tensor.quantization_info());
130                 }
131                 break;
132             case DataType::QASYMM16:
133                 for(size_t i = 0; i < values.size(); ++i)
134                 {
135                     data_ptr[i] = quantize_qasymm16(values[i], tensor.quantization_info());
136                 }
137                 break;
138             default:
139                 for(size_t i = 0; i < values.size(); ++i)
140                 {
141                     data_ptr[i] = static_cast<data_type>(values[i]);
142                 }
143         }
144     }
145 
compute_target(const TensorShape & deltas_shape,DataType data_type,const BoundingBoxTransformInfo & bbox_info,std::mt19937 & gen,QuantizationInfo deltas_qinfo)146     TensorType compute_target(const TensorShape &deltas_shape, DataType data_type,
147                               const BoundingBoxTransformInfo &bbox_info, std::mt19937 &gen,
148                               QuantizationInfo deltas_qinfo)
149     {
150         // Create tensors
151         TensorShape boxes_shape(4, deltas_shape[1]);
152         TensorType  deltas = create_tensor<TensorType>(deltas_shape, _data_type_deltas, 1, deltas_qinfo);
153         TensorType  boxes  = create_tensor<TensorType>(boxes_shape, data_type, 1, _boxes_qinfo);
154         TensorType  pred_boxes;
155 
156         // Create and configure function
157         FunctionType bbox_transform;
158         bbox_transform.configure(&boxes, &pred_boxes, &deltas, bbox_info);
159 
160         ARM_COMPUTE_ASSERT(deltas.info()->is_resizable());
161         ARM_COMPUTE_ASSERT(boxes.info()->is_resizable());
162         ARM_COMPUTE_ASSERT(pred_boxes.info()->is_resizable());
163 
164         // Allocate tensors
165         deltas.allocator()->allocate();
166         boxes.allocator()->allocate();
167         pred_boxes.allocator()->allocate();
168 
169         ARM_COMPUTE_ASSERT(!deltas.info()->is_resizable());
170         ARM_COMPUTE_ASSERT(!boxes.info()->is_resizable());
171 
172         // Fill tensors
173         TensorShape        img_shape(bbox_info.scale() * bbox_info.img_width(), bbox_info.scale() * bbox_info.img_height());
174         std::vector<float> boxes_vec  = generate_boxes(img_shape, boxes_shape[1], gen);
175         std::vector<float> deltas_vec = generate_deltas(boxes_vec, img_shape, deltas_shape[1], deltas_shape[0] / 4, gen);
176         fill<T>(AccessorType(boxes), boxes_vec);
177         fill<TDeltas>(AccessorType(deltas), deltas_vec);
178 
179         // Compute function
180         bbox_transform.run();
181 
182         return pred_boxes;
183     }
184 
compute_reference(const TensorShape & deltas_shape,DataType data_type,const BoundingBoxTransformInfo & bbox_info,std::mt19937 & gen,QuantizationInfo deltas_qinfo)185     SimpleTensor<T> compute_reference(const TensorShape              &deltas_shape,
186                                       DataType                        data_type,
187                                       const BoundingBoxTransformInfo &bbox_info,
188                                       std::mt19937                   &gen,
189                                       QuantizationInfo                deltas_qinfo)
190     {
191         // Create reference tensor
192         TensorShape           boxes_shape(4, deltas_shape[1]);
193         SimpleTensor<T>       boxes{ boxes_shape, data_type, 1, _boxes_qinfo };
194         SimpleTensor<TDeltas> deltas{ deltas_shape, _data_type_deltas, 1, deltas_qinfo };
195 
196         // Fill reference tensor
197         TensorShape        img_shape(bbox_info.scale() * bbox_info.img_width(), bbox_info.scale() * bbox_info.img_height());
198         std::vector<float> boxes_vec  = generate_boxes(img_shape, boxes_shape[1], gen);
199         std::vector<float> deltas_vec = generate_deltas(boxes_vec, img_shape, deltas_shape[1], deltas_shape[0] / 4, gen);
200         fill<T>(boxes, boxes_vec);
201         fill<TDeltas>(deltas, deltas_vec);
202 
203         return reference::bounding_box_transform(boxes, deltas, bbox_info);
204     }
205 
206     TensorType       _target{};
207     SimpleTensor<T>  _reference{};
208     DataType         _data_type_deltas{};
209     QuantizationInfo _boxes_qinfo{};
210 
211 private:
212 };
213 
214 template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
215 class BoundingBoxTransformFixture : public BoundingBoxTransformGenericFixture<TensorType, AccessorType, FunctionType, T>
216 {
217 public:
218     template <typename...>
setup(TensorShape deltas_shape,const BoundingBoxTransformInfo & info,DataType data_type)219     void setup(TensorShape deltas_shape, const BoundingBoxTransformInfo &info, DataType data_type)
220     {
221         BoundingBoxTransformGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(deltas_shape, info, data_type, QuantizationInfo());
222     }
223 
224 private:
225 };
226 
227 template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
228 class BoundingBoxTransformQuantizedFixture : public BoundingBoxTransformGenericFixture<TensorType, AccessorType, FunctionType, T>
229 {
230 public:
231     template <typename...>
setup(TensorShape deltas_shape,const BoundingBoxTransformInfo & info,DataType data_type,QuantizationInfo deltas_qinfo)232     void setup(TensorShape deltas_shape, const BoundingBoxTransformInfo &info, DataType data_type, QuantizationInfo deltas_qinfo)
233     {
234         BoundingBoxTransformGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(deltas_shape, info, data_type, deltas_qinfo);
235     }
236 };
237 } // namespace validation
238 } // namespace test
239 } // namespace arm_compute
240 #endif /* ARM_COMPUTE_TEST_BOUNDINGBOXTRANSFORM_FIXTURE */
241