Gitiles
Code Review Sign In
review.mlplatform.org / ml / ComputeLibrary / d64a46c6dfa81ce4607fc3de57bc9d9ac7e01e4a / . / tests / validation / NEON / GenerateProposalsLayer.cpp
blob: ea99bb310725ff05d72b9144a0a0477a16a28702 [file] [log] [blame]
Pablo Telloc9564cb2019-09-13 10:20:25 +0100[diff] [blame]1/*
2 * Copyright (c) 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 */
24#include "arm_compute/runtime/NEON/NEScheduler.h"
25#include "arm_compute/runtime/NEON/functions/NEComputeAllAnchors.h"
26#include "arm_compute/runtime/NEON/functions/NEGenerateProposalsLayer.h"
27#include "arm_compute/runtime/NEON/functions/NEPermute.h"
28#include "arm_compute/runtime/NEON/functions/NESlice.h"
29#include "tests/Globals.h"
30#include "tests/NEON/Accessor.h"
31#include "tests/NEON/ArrayAccessor.h"
32#include "tests/framework/Macros.h"
33#include "tests/framework/datasets/Datasets.h"
34#include "tests/validation/Validation.h"
35#include "tests/validation/fixtures/ComputeAllAnchorsFixture.h"
36#include "utils/TypePrinter.h"
37
38namespace arm_compute
39{
40namespace test
41{
42namespace validation
43{
44namespace
45{
46template <typename U, typename T>
47inline void fill_tensor(U &&tensor, const std::vector<T> &v)
48{
49 std::memcpy(tensor.data(), v.data(), sizeof(T) * v.size());
50}
51
52template <typename T>
53inline void fill_tensor(Accessor &&tensor, const std::vector<T> &v)
54{
55 if(tensor.data_layout() == DataLayout::NCHW)
56 {
57 std::memcpy(tensor.data(), v.data(), sizeof(T) * v.size());
58 }
59 else
60 {
61 const int channels = tensor.shape()[0];
62 const int width = tensor.shape()[1];
63 const int height = tensor.shape()[2];
64 for(int x = 0; x < width; ++x)
65 {
66 for(int y = 0; y < height; ++y)
67 {
68 for(int c = 0; c < channels; ++c)
69 {
70 *(reinterpret_cast<T *>(tensor(Coordinates(c, x, y)))) = *(reinterpret_cast<const T *>(v.data() + x + y * width + c * height * width));
71 }
72 }
73 }
74 }
75}
76
77const auto ComputeAllInfoDataset = framework::dataset::make("ComputeAllInfo",
78{
79 ComputeAnchorsInfo(10U, 10U, 1. / 16.f),
80 ComputeAnchorsInfo(100U, 1U, 1. / 2.f),
81 ComputeAnchorsInfo(100U, 1U, 1. / 4.f),
82 ComputeAnchorsInfo(100U, 100U, 1. / 4.f),
83
84});
85} // namespace
86
87TEST_SUITE(NEON)
88TEST_SUITE(GenerateProposals)
89
90// *INDENT-OFF*
91// clang-format off
92DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zip(
93 framework::dataset::make("scores", { TensorInfo(TensorShape(100U, 100U, 9U), 1, DataType::F32),
94 TensorInfo(TensorShape(100U, 100U, 9U), 1, DataType::F16), // Mismatching types
95 TensorInfo(TensorShape(100U, 100U, 9U), 1, DataType::F16), // Wrong deltas (number of transformation non multiple of 4)
96 TensorInfo(TensorShape(100U, 100U, 9U), 1, DataType::F16), // Wrong anchors (number of values per roi != 5)
97 TensorInfo(TensorShape(100U, 100U, 9U), 1, DataType::F16), // Output tensor num_valid_proposals not scalar
98 TensorInfo(TensorShape(100U, 100U, 9U), 1, DataType::F16)}), // num_valid_proposals not U32
99 framework::dataset::make("deltas",{ TensorInfo(TensorShape(100U, 100U, 36U), 1, DataType::F32),
100 TensorInfo(TensorShape(100U, 100U, 36U), 1, DataType::F32),
101 TensorInfo(TensorShape(100U, 100U, 38U), 1, DataType::F32),
102 TensorInfo(TensorShape(100U, 100U, 38U), 1, DataType::F32),
103 TensorInfo(TensorShape(100U, 100U, 38U), 1, DataType::F32),
104 TensorInfo(TensorShape(100U, 100U, 38U), 1, DataType::F32)})),
105 framework::dataset::make("anchors", { TensorInfo(TensorShape(4U, 9U), 1, DataType::F32),
106 TensorInfo(TensorShape(4U, 9U), 1, DataType::F32),
107 TensorInfo(TensorShape(4U, 9U), 1, DataType::F32),
108 TensorInfo(TensorShape(5U, 9U), 1, DataType::F32),
109 TensorInfo(TensorShape(4U, 9U), 1, DataType::F32),
110 TensorInfo(TensorShape(4U, 9U), 1, DataType::F32)})),
111 framework::dataset::make("proposals", { TensorInfo(TensorShape(5U, 100U*100U*9U), 1, DataType::F32),
112 TensorInfo(TensorShape(5U, 100U*100U*9U), 1, DataType::F32),
113 TensorInfo(TensorShape(5U, 100U*100U*9U), 1, DataType::F32),
114 TensorInfo(TensorShape(5U, 100U*100U*9U), 1, DataType::F32),
115 TensorInfo(TensorShape(5U, 100U*100U*9U), 1, DataType::F32),
116 TensorInfo(TensorShape(5U, 100U*100U*9U), 1, DataType::F32)})),
117 framework::dataset::make("scores_out", { TensorInfo(TensorShape(100U*100U*9U), 1, DataType::F32),
118 TensorInfo(TensorShape(100U*100U*9U), 1, DataType::F32),
119 TensorInfo(TensorShape(100U*100U*9U), 1, DataType::F32),
120 TensorInfo(TensorShape(100U*100U*9U), 1, DataType::F32),
121 TensorInfo(TensorShape(100U*100U*9U), 1, DataType::F32),
122 TensorInfo(TensorShape(100U*100U*9U), 1, DataType::F32)})),
123 framework::dataset::make("num_valid_proposals", { TensorInfo(TensorShape(1U, 1U), 1, DataType::U32),
124 TensorInfo(TensorShape(1U, 1U), 1, DataType::U32),
125 TensorInfo(TensorShape(1U, 1U), 1, DataType::U32),
126 TensorInfo(TensorShape(1U, 1U), 1, DataType::U32),
127 TensorInfo(TensorShape(1U, 10U), 1, DataType::U32),
128 TensorInfo(TensorShape(1U, 1U), 1, DataType::F16)})),
129 framework::dataset::make("generate_proposals_info", { GenerateProposalsInfo(10.f, 10.f, 1.f),
130 GenerateProposalsInfo(10.f, 10.f, 1.f),
131 GenerateProposalsInfo(10.f, 10.f, 1.f),
132 GenerateProposalsInfo(10.f, 10.f, 1.f),
133 GenerateProposalsInfo(10.f, 10.f, 1.f),
134 GenerateProposalsInfo(10.f, 10.f, 1.f)})),
135 framework::dataset::make("Expected", { true, false, false, false, false, false })),
136 scores, deltas, anchors, proposals, scores_out, num_valid_proposals, generate_proposals_info, expected)
137{
138 ARM_COMPUTE_EXPECT(bool(NEGenerateProposalsLayer::validate(&scores.clone()->set_is_resizable(true),
139 &deltas.clone()->set_is_resizable(true),
140 &anchors.clone()->set_is_resizable(true),
141 &proposals.clone()->set_is_resizable(true),
142 &scores_out.clone()->set_is_resizable(true),
143 &num_valid_proposals.clone()->set_is_resizable(true),
144 generate_proposals_info)) == expected, framework::LogLevel::ERRORS);
145}
146// clang-format on
147// *INDENT-ON*
148
149template <typename T>
150using NEComputeAllAnchorsFixture = ComputeAllAnchorsFixture<Tensor, Accessor, NEComputeAllAnchors, T>;
151
152TEST_SUITE(Float)
153TEST_SUITE(FP32)
154DATA_TEST_CASE(IntegrationTestCaseAllAnchors, framework::DatasetMode::ALL, framework::dataset::make("DataType", { DataType::F32 }),
155 data_type)
156{
157 const int values_per_roi = 4;
158 const int num_anchors = 3;
159 const int feature_height = 4;
160 const int feature_width = 3;
161
162 SimpleTensor<float> anchors_expected(TensorShape(values_per_roi, feature_width * feature_height * num_anchors), DataType::F32);
163 fill_tensor(anchors_expected, std::vector<float> { -26, -19, 87, 86,
164 -81, -27, 58, 63,
165 -44, -15, 55, 36,
166 -10, -19, 103, 86,
167 -65, -27, 74, 63,
168 -28, -15, 71, 36,
169 6, -19, 119, 86,
170 -49, -27, 90, 63,
171 -12, -15, 87, 36,
172 -26, -3, 87, 102,
173 -81, -11, 58, 79,
174 -44, 1, 55, 52,
175 -10, -3, 103, 102,
176 -65, -11, 74, 79,
177 -28, 1, 71, 52,
178 6, -3, 119, 102,
179 -49, -11, 90, 79,
180 -12, 1, 87, 52,
181 -26, 13, 87, 118,
182 -81, 5, 58, 95,
183 -44, 17, 55, 68,
184 -10, 13, 103, 118,
185 -65, 5, 74, 95,
186 -28, 17, 71, 68,
187 6, 13, 119, 118,
188 -49, 5, 90, 95,
189 -12, 17, 87, 68,
190 -26, 29, 87, 134,
191 -81, 21, 58, 111,
192 -44, 33, 55, 84,
193 -10, 29, 103, 134,
194 -65, 21, 74, 111,
195 -28, 33, 71, 84,
196 6, 29, 119, 134,
197 -49, 21, 90, 111,
198 -12, 33, 87, 84
199 });
200
201 Tensor all_anchors;
202 Tensor anchors = create_tensor<Tensor>(TensorShape(4, num_anchors), data_type);
203
204 // Create and configure function
205 NEComputeAllAnchors compute_anchors;
206 compute_anchors.configure(&anchors, &all_anchors, ComputeAnchorsInfo(feature_width, feature_height, 1. / 16.0));
207 anchors.allocator()->allocate();
208 all_anchors.allocator()->allocate();
209
210 fill_tensor(Accessor(anchors), std::vector<float> { -26, -19, 87, 86,
211 -81, -27, 58, 63,
212 -44, -15, 55, 36
213 });
214 // Compute function
215 compute_anchors.run();
216 validate(Accessor(all_anchors), anchors_expected);
217}
218
219DATA_TEST_CASE(IntegrationTestCaseGenerateProposals, framework::DatasetMode::ALL, combine(framework::dataset::make("DataType", { DataType::F32 }),
220 framework::dataset::make("DataLayout", { DataLayout::NCHW, DataLayout::NHWC })),
221 data_type, data_layout)
222{
223 const int values_per_roi = 4;
224 const int num_anchors = 2;
225 const int feature_height = 4;
226 const int feature_width = 5;
227
228 std::vector<float> scores_vector
229 {
230 5.055894435664012e-04f, 1.270304909820112e-03f, 2.492271113912067e-03f, 5.951663827809190e-03f,
231 7.846917156877404e-03f, 6.776275276294789e-03f, 6.761571012891965e-03f, 4.898292096237725e-03f,
232 6.044472332578605e-04f, 3.203334118759474e-03f, 2.947527908919908e-03f, 6.313238560015770e-03f,
233 7.931767757095738e-03f, 8.764345805102866e-03f, 7.325012199914913e-03f, 4.317069470446271e-03f,
234 2.372537409795522e-03f, 1.589227460352735e-03f, 7.419477503600818e-03f, 3.157690354133824e-05f,
235 1.125915135986472e-03f, 9.865363483872330e-03f, 2.429454743386769e-03f, 2.724460564167563e-03f,
236 7.670409838207963e-03f, 5.558891552328172e-03f, 7.876904873099614e-03f, 6.824746047239291e-03f,
237 7.023817548067892e-03f, 3.651314909238673e-04f, 6.720443709032501e-03f, 5.935615511606155e-03f,
238 2.837349642759774e-03f, 1.787235113610299e-03f, 4.538568889918262e-03f, 3.391510678188818e-03f,
239 7.328474239481874e-03f, 6.306967923936016e-03f, 8.102218904895860e-04f, 3.366646521610209e-03f
240 };
241
242 std::vector<float> bbx_vector
243 {
244 5.066650471856862e-03, -7.638671742936328e-03, 2.549596503988635e-03, -8.316416756423296e-03,
245 -2.397471917924575e-04, 7.370595187754891e-03, -2.771880178185262e-03, 3.958364873973579e-03,
246 4.493661094712284e-03, 2.016487051533088e-03, -5.893883038142033e-03, 7.570636080807809e-03,
247 -1.395511229386785e-03, 3.686686052704696e-03, -7.738166245767079e-03, -1.947306329828059e-03,
248 -9.299719716045681e-03, -3.476410493413708e-03, -2.390761190919604e-03, 4.359281254364210e-03,
249 -2.135251160164030e-04, 9.203299843371962e-03, 4.042322775006053e-03, -9.464271243910754e-03,
250 2.566239543229305e-03, -9.691093900220627e-03, -4.019283034310979e-03, 8.145470429508792e-03,
251 7.345087308315662e-04, 7.049642787384043e-03, -2.768492313674294e-03, 6.997160053405803e-03,
252 6.675346697112969e-03, 2.353293365652274e-03, -3.612002585241749e-04, 1.592076522068768e-03,
253 -8.354188900818149e-04, -5.232515333564140e-04, 6.946683728847089e-03, -8.469757407935994e-03,
254 -8.985324496496555e-03, 4.885832859017961e-03, -7.662967577576512e-03, 7.284124004335807e-03,
255 -5.812167510299458e-03, -5.760336800482398e-03, 6.040416930336549e-03, 5.861508595443691e-03,
256 -5.509243096133549e-04, -2.006142470055888e-03, -7.205925340416066e-03, -1.117459082969758e-03,
257 4.233247017623154e-03, 8.079257498201178e-03, 2.962639022639513e-03, 7.069474943472751e-03,
258 -8.562946284971293e-03, -8.228634642768271e-03, -6.116245322799971e-04, -7.213122000180859e-03,
259 1.693094399433209e-03, -4.287504459132290e-03, 8.740365683925144e-03, 3.751788160720638e-03,
260 7.006764222862830e-03, 9.676754678358187e-03, -6.458757235812945e-03, -4.486506575589758e-03,
261 -4.371087196816259e-03, 3.542166755953152e-03, -2.504808998699504e-03, 5.666601724512010e-03,
262 -3.691862724546129e-03, 3.689809719085287e-03, 9.079930264704458e-03, 6.365127787359476e-03,
263 2.881681788246101e-06, 9.991866069315165e-03, -1.104757466496565e-03, -2.668455405633477e-03,
264 -1.225748887087659e-03, 6.530536159094015e-03, 3.629468917975644e-03, 1.374426066950348e-03,
265 -2.404098881570632e-03, -4.791365049441602e-03, -2.970654027009094e-03, 7.807553690294366e-03,
266 -1.198321129505323e-03, -3.574885336949881e-03, -5.380848303732298e-03, 9.705151282165116e-03,
267 -1.005217683242201e-03, 9.178094036278405e-03, -5.615977269541644e-03, 5.333533158509859e-03,
268 -2.817116206168516e-03, 6.672609782000503e-03, 6.575769501651313e-03, 8.987596634989362e-03,
269 -1.283530791296188e-03, 1.687717120057778e-03, 3.242391851439037e-03, -7.312060454341677e-03,
270 4.735335326324270e-03, -6.832367028817463e-03, -5.414854835884652e-03, -9.352380213755996e-03,
271 -3.682662043703889e-03, -6.127508590419776e-04, -7.682256596819467e-03, 9.569532628790246e-03,
272 -1.572157284518933e-03, -6.023034366859191e-03, -5.110873282582924e-03, -8.697072236660256e-03,
273 -3.235150419663566e-03, -8.286320236471386e-03, -5.229472409112913e-03, 9.920785896115053e-03,
274 -2.478413362126123e-03, -9.261324796935007e-03, 1.718512310840434e-04, 3.015875488208480e-03,
275 -6.172932549255669e-03, -4.031715551985103e-03, -9.263878005853677e-03, -2.815310738453385e-03,
276 7.075307462133643e-03, 1.404611747938669e-03, -1.518548732533266e-03, -9.293430941655778e-03,
277 6.382186966633246e-03, 8.256835789169248e-03, 3.196907843506736e-03, 8.821615689753433e-03,
278 -7.661543424832439e-03, 1.636273081822326e-03, -8.792373335756125e-03, 2.958775812049877e-03,
279 -6.269300278071262e-03, 6.248285790856450e-03, -3.675414624536002e-03, -1.692616700318762e-03,
280 4.126007647815893e-03, -9.155291689759584e-03, -8.432616039924004e-03, 4.899980636213323e-03,
281 3.511535019681671e-03, -1.582745757177339e-03, -2.703657774917963e-03, 6.738168990840388e-03,
282 4.300455303937919e-03, 9.618312854781494e-03, 2.762142918402472e-03, -6.590025003382154e-03,
283 -2.071168373801788e-03, 8.613893943683627e-03, 9.411190295341036e-03, -6.129018930548372e-03
284 };
285
286 const std::vector<float> anchors_vector{ -26, -19, 87, 86, -81, -27, 58, 63 };
287 ;
288
289 SimpleTensor<float> proposals_expected(TensorShape(5, 9), DataType::F32);
290 fill_tensor(proposals_expected, std::vector<float>
291 {
292 0, 0, 0, 75.269, 64.4388,
293 0, 21.9579, 13.0535, 119, 99,
294 0, 38.303, 0, 119, 87.6447,
295 0, 0, 0, 119, 64.619,
296 0, 0, 20.7997, 74.0714, 99,
297 0, 0, 0, 91.8963, 79.3724,
298 0, 0, 4.42377, 58.1405, 95.1781,
299 0, 0, 13.4405, 104.799, 99,
300 0, 38.9066, 28.2434, 119, 99,
301
302 });
303
304 SimpleTensor<float> scores_expected(TensorShape(9), DataType::F32);
305 fill_tensor(scores_expected, std::vector<float>
306 {
307 0.00986536,
308 0.00876435,
309 0.00784692,
310 0.00767041,
311 0.00732847,
312 0.00682475,
313 0.00672044,
314 0.00631324,
315 3.15769e-05
316 });
317
318 TensorShape scores_shape = TensorShape(feature_width, feature_height, num_anchors);
319 TensorShape deltas_shape = TensorShape(feature_width, feature_height, values_per_roi * num_anchors);
320 if(data_layout == DataLayout::NHWC)
321 {
322 permute(scores_shape, PermutationVector(2U, 0U, 1U));
323 permute(deltas_shape, PermutationVector(2U, 0U, 1U));
324 }
325 // Inputs
326 Tensor scores = create_tensor<Tensor>(scores_shape, data_type, 1, QuantizationInfo(), data_layout);
327 Tensor bbox_deltas = create_tensor<Tensor>(deltas_shape, data_type, 1, QuantizationInfo(), data_layout);
328 Tensor anchors = create_tensor<Tensor>(TensorShape(values_per_roi, num_anchors), data_type);
329
330 // Outputs
331 Tensor proposals;
332 Tensor num_valid_proposals;
333 Tensor scores_out;
334 num_valid_proposals.allocator()->init(TensorInfo(TensorShape(1), 1, DataType::U32));
335
336 NEGenerateProposalsLayer generate_proposals;
337 generate_proposals.configure(&scores, &bbox_deltas, &anchors, &proposals, &scores_out, &num_valid_proposals,
338 GenerateProposalsInfo(120, 100, 0.166667f, 1 / 16.0, 6000, 300, 0.7f, 16.0f));
339
340 // Allocate memory for input/output tensors
341 scores.allocator()->allocate();
342 bbox_deltas.allocator()->allocate();
343 anchors.allocator()->allocate();
344 proposals.allocator()->allocate();
345 num_valid_proposals.allocator()->allocate();
346 scores_out.allocator()->allocate();
347 // Fill inputs
348 fill_tensor(Accessor(scores), scores_vector);
349 fill_tensor(Accessor(bbox_deltas), bbx_vector);
350 fill_tensor(Accessor(anchors), anchors_vector);
351
352 // Run operator
353 generate_proposals.run();
354 // Gather num_valid_proposals
355 const uint32_t N = *reinterpret_cast<uint32_t *>(num_valid_proposals.ptr_to_element(Coordinates(0, 0)));
356
357 // Select the first N entries of the proposals
358 Tensor proposals_final;
359 NESlice select_proposals;
360 select_proposals.configure(&proposals, &proposals_final, Coordinates(0, 0), Coordinates(values_per_roi + 1, N));
361
362 proposals_final.allocator()->allocate();
363 select_proposals.run();
364
365 // Select the first N entries of the proposals
366 Tensor scores_final;
367 NESlice select_scores;
368 select_scores.configure(&scores_out, &scores_final, Coordinates(0), Coordinates(N));
369 scores_final.allocator()->allocate();
370 select_scores.run();
371
372 const RelativeTolerance<float> tolerance_f32(1e-5f);
373 // Validate the output
374 validate(Accessor(proposals_final), proposals_expected, tolerance_f32);
375 validate(Accessor(scores_final), scores_expected, tolerance_f32);
376}
377
378FIXTURE_DATA_TEST_CASE(ComputeAllAnchors, NEComputeAllAnchorsFixture<float>, framework::DatasetMode::ALL,
379 combine(combine(framework::dataset::make("NumAnchors", { 2, 4, 8 }), ComputeAllInfoDataset), framework::dataset::make("DataType", { DataType::F32 })))
380{
381 // Validate output
382 validate(Accessor(_target), _reference);
383}
384TEST_SUITE_END() // FP32
385#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
386TEST_SUITE(FP16)
387FIXTURE_DATA_TEST_CASE(ComputeAllAnchors, NEComputeAllAnchorsFixture<half>, framework::DatasetMode::ALL,
388 combine(combine(framework::dataset::make("NumAnchors", { 2, 4, 8 }), ComputeAllInfoDataset), framework::dataset::make("DataType", { DataType::F16 })))
389{
390 // Validate output
391 validate(Accessor(_target), _reference);
392}
393TEST_SUITE_END() // FP16
394#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
395
396TEST_SUITE_END() // Float
397
398TEST_SUITE_END() // GenerateProposals
399TEST_SUITE_END() // NEON
400
401} // namespace validation
402} // namespace test
403} // namespace arm_compute