John Richardson | 7f4a819 | 2018-02-05 15:12:22 +0000 | [diff] [blame] | 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 | #ifndef ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_FIXTURE |
| 25 | #define ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_FIXTURE |
| 26 | |
| 27 | #include "arm_compute/core/HOGInfo.h" |
| 28 | #include "arm_compute/core/TensorShape.h" |
| 29 | #include "arm_compute/core/Types.h" |
| 30 | #include "tests/AssetsLibrary.h" |
| 31 | #include "tests/Globals.h" |
| 32 | #include "tests/IAccessor.h" |
| 33 | #include "tests/IHOGAccessor.h" |
| 34 | #include "tests/framework/Asserts.h" |
| 35 | #include "tests/framework/Fixture.h" |
| 36 | #include "tests/validation/reference/HOGMultiDetection.h" |
| 37 | |
| 38 | namespace arm_compute |
| 39 | { |
| 40 | namespace test |
| 41 | { |
| 42 | namespace validation |
| 43 | { |
| 44 | template <typename TensorType, |
| 45 | typename HOGType, |
| 46 | typename MultiHOGType, |
| 47 | typename DetectionWindowArrayType, |
| 48 | typename DetectionWindowStrideType, |
| 49 | typename AccessorType, |
| 50 | typename Size2DArrayAccessorType, |
| 51 | typename DetectionWindowArrayAccessorType, |
| 52 | typename HOGAccessorType, |
| 53 | typename FunctionType, |
| 54 | typename T, |
| 55 | typename U> |
| 56 | class HOGMultiDetectionValidationFixture : public framework::Fixture |
| 57 | { |
| 58 | public: |
| 59 | template <typename...> |
| 60 | void setup(std::string image, std::vector<HOGInfo> models, Format format, BorderMode border_mode, bool non_maxima_suppression) |
| 61 | { |
| 62 | // Only defined borders supported |
| 63 | ARM_COMPUTE_ERROR_ON(border_mode == BorderMode::UNDEFINED); |
| 64 | |
| 65 | // Generate a random constant value |
| 66 | std::mt19937 gen(library->seed()); |
| 67 | std::uniform_int_distribution<T> int_dist(0, 255); |
| 68 | const T constant_border_value = int_dist(gen); |
| 69 | |
| 70 | // Initialize descriptors vector |
| 71 | std::vector<std::vector<U>> descriptors(models.size()); |
| 72 | |
| 73 | // Use default values for threshold and min_distance |
| 74 | const float threshold = 0.f; |
| 75 | const float min_distance = 1.f; |
| 76 | |
| 77 | // Maximum number of detection windows per batch |
| 78 | const unsigned int max_num_detection_windows = 100000; |
| 79 | |
| 80 | _target = compute_target(image, format, border_mode, constant_border_value, models, descriptors, max_num_detection_windows, threshold, non_maxima_suppression, min_distance); |
| 81 | _reference = compute_reference(image, format, border_mode, constant_border_value, models, descriptors, max_num_detection_windows, threshold, non_maxima_suppression, min_distance); |
| 82 | } |
| 83 | |
| 84 | protected: |
| 85 | template <typename V> |
| 86 | void fill(V &&tensor, const std::string image, Format format) |
| 87 | { |
| 88 | library->fill(tensor, image, format); |
| 89 | } |
| 90 | |
| 91 | void initialize_batch(const std::vector<HOGInfo> &models, MultiHOGType &multi_hog, |
| 92 | std::vector<std::vector<U>> &descriptors, DetectionWindowStrideType &detection_window_strides) |
| 93 | { |
| 94 | for(unsigned i = 0; i < models.size(); ++i) |
| 95 | { |
| 96 | auto hog_model = reinterpret_cast<HOGType *>(multi_hog.model(i)); |
| 97 | hog_model->init(models[i]); |
| 98 | |
| 99 | // Initialise descriptor (linear SVM coefficients). |
| 100 | std::random_device::result_type seed = 0; |
| 101 | descriptors.at(i) = generate_random_real(models[i].descriptor_size(), -0.505f, 0.495f, seed); |
| 102 | |
| 103 | // Copy HOG descriptor values to HOG memory |
| 104 | { |
| 105 | HOGAccessorType hog_accessor(*hog_model); |
| 106 | std::memcpy(hog_accessor.descriptor(), descriptors.at(i).data(), descriptors.at(i).size() * sizeof(U)); |
| 107 | } |
| 108 | |
| 109 | // Initialize detection window stride |
| 110 | Size2DArrayAccessorType accessor(detection_window_strides); |
| 111 | accessor.at(i) = models[i].block_stride(); |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | std::vector<DetectionWindow> compute_target(const std::string image, Format &format, BorderMode &border_mode, T constant_border_value, |
| 116 | const std::vector<HOGInfo> &models, std::vector<std::vector<U>> &descriptors, unsigned int max_num_detection_windows, |
| 117 | float threshold, bool non_max_suppression, float min_distance) |
| 118 | { |
| 119 | MultiHOGType multi_hog(models.size()); |
| 120 | DetectionWindowArrayType detection_windows(max_num_detection_windows); |
| 121 | DetectionWindowStrideType detection_window_strides(models.size()); |
| 122 | |
| 123 | // Resize detection window_strides for index access |
| 124 | detection_window_strides.resize(models.size()); |
| 125 | |
| 126 | // Initialiize MultiHOG and detection windows |
| 127 | initialize_batch(models, multi_hog, descriptors, detection_window_strides); |
| 128 | |
| 129 | // Get image shape for src tensor |
| 130 | TensorShape shape = library->get_image_shape(image); |
| 131 | |
| 132 | // Create tensors |
| 133 | TensorType src = create_tensor<TensorType>(shape, data_type_from_format(format)); |
| 134 | ARM_COMPUTE_EXPECT(src.info()->is_resizable(), framework::LogLevel::ERRORS); |
| 135 | |
| 136 | // Create and configure function |
| 137 | FunctionType hog_multi_detection; |
| 138 | hog_multi_detection.configure(&src, &multi_hog, &detection_windows, &detection_window_strides, border_mode, constant_border_value, threshold, non_max_suppression, min_distance); |
| 139 | |
| 140 | // Reset detection windows |
| 141 | detection_windows.clear(); |
| 142 | |
| 143 | // Allocate tensors |
| 144 | src.allocator()->allocate(); |
| 145 | ARM_COMPUTE_EXPECT(!src.info()->is_resizable(), framework::LogLevel::ERRORS); |
| 146 | |
| 147 | // Fill tensors |
| 148 | fill(AccessorType(src), image, format); |
| 149 | |
| 150 | // Compute function |
| 151 | hog_multi_detection.run(); |
| 152 | |
| 153 | // Copy detection windows |
| 154 | std::vector<DetectionWindow> windows; |
| 155 | DetectionWindowArrayAccessorType accessor(detection_windows); |
| 156 | |
| 157 | for(size_t i = 0; i < accessor.num_values(); i++) |
| 158 | { |
| 159 | DetectionWindow win; |
| 160 | win.x = accessor.at(i).x; |
| 161 | win.y = accessor.at(i).y; |
| 162 | win.width = accessor.at(i).width; |
| 163 | win.height = accessor.at(i).height; |
| 164 | win.idx_class = accessor.at(i).idx_class; |
| 165 | win.score = accessor.at(i).score; |
| 166 | |
| 167 | windows.push_back(win); |
| 168 | } |
| 169 | |
| 170 | return windows; |
| 171 | } |
| 172 | |
| 173 | std::vector<DetectionWindow> compute_reference(const std::string image, Format format, BorderMode border_mode, T constant_border_value, |
| 174 | const std::vector<HOGInfo> &models, const std::vector<std::vector<U>> &descriptors, unsigned int max_num_detection_windows, |
| 175 | float threshold, bool non_max_suppression, float min_distance) |
| 176 | { |
| 177 | // Create reference |
| 178 | SimpleTensor<T> src{ library->get_image_shape(image), data_type_from_format(format) }; |
| 179 | |
| 180 | // Fill reference |
| 181 | fill(src, image, format); |
| 182 | |
| 183 | // NOTE: Detection window stride fixed to block stride |
| 184 | return reference::hog_multi_detection(src, border_mode, constant_border_value, models, descriptors, max_num_detection_windows, threshold, non_max_suppression, min_distance); |
| 185 | } |
| 186 | |
| 187 | std::vector<DetectionWindow> _target{}; |
| 188 | std::vector<DetectionWindow> _reference{}; |
| 189 | }; |
| 190 | } // namespace validation |
| 191 | } // namespace test |
| 192 | } // namespace arm_compute |
| 193 | #endif /* ARM_COMPUTE_TEST_HOG_MULTI_DETECTION_FIXTURE */ |