ncnn

Форк
0
/
test_deconvolution.cpp 
244 строки · 9.2 Кб
1
// Tencent is pleased to support the open source community by making ncnn available.
2
//
3
// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
4
//
5
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
6
// in compliance with the License. You may obtain a copy of the License at
7
//
8
// https://opensource.org/licenses/BSD-3-Clause
9
//
10
// Unless required by applicable law or agreed to in writing, software distributed
11
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
12
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
13
// specific language governing permissions and limitations under the License.
14

15
#include "testutil.h"
16

17
static int test_deconvolution(int w, int h, int c, int outch, int kernel, int dilation, int stride, int pad, int bias, int output_pad_right, int output_pad_bottom, int output_w, int output_h)
18
{
19
    ncnn::Mat a = RandomMat(w, h, c);
20

21
    if (output_w > 0 && output_h > 0 && pad != -233 && pad != -234)
22
    {
23
        pad = -233;
24
    }
25

26
    ncnn::ParamDict pd;
27
    pd.set(0, outch);    // num_output
28
    pd.set(1, kernel);   // kernel_w
29
    pd.set(2, dilation); // dilation_w
30
    pd.set(3, stride);   // stride_w
31
    pd.set(4, pad);      // pad_w
32
    pd.set(5, bias);     // bias_term
33
    pd.set(6, outch * c * kernel * kernel);
34

35
    int activation_type = RAND() % 5; // 0 1 2 3 4
36
    ncnn::Mat activation_params(2);
37
    activation_params[0] = RandomFloat(-1, 0); // alpha
38
    activation_params[1] = RandomFloat(0, 1);  // beta
39
    pd.set(9, activation_type);
40
    pd.set(10, activation_params);
41

42
    pd.set(18, output_pad_right);
43
    pd.set(19, output_pad_bottom);
44
    pd.set(20, output_w);
45
    pd.set(21, output_h);
46

47
    std::vector<ncnn::Mat> weights(2);
48
    weights[0] = RandomMat(outch * c * kernel * kernel);
49
    weights[1] = RandomMat(outch);
50

51
    int ret = test_layer("Deconvolution", pd, weights, a);
52
    if (ret != 0)
53
    {
54
        fprintf(stderr, "test_deconvolution failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f] output_pad_right=%d output_pad_bottom=%d output_w=%d output_h=%d\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1], output_pad_right, output_pad_bottom, output_w, output_h);
55
    }
56

57
    {
58
        ncnn::Option opt;
59
        opt.num_threads = 1;
60
        opt.use_packing_layout = true;
61
        opt.use_fp16_packed = false;
62
        opt.use_fp16_storage = false;
63
        opt.use_fp16_arithmetic = false;
64
        opt.use_bf16_storage = false;
65
        opt.use_shader_pack8 = false;
66
        opt.use_image_storage = false;
67
        opt.use_sgemm_convolution = false;
68
        opt.use_winograd_convolution = false;
69

70
        ret = test_layer_opt("Deconvolution", pd, weights, opt, a);
71
        if (ret != 0)
72
        {
73
            fprintf(stderr, "test_deconvolution failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f] output_pad_right=%d output_pad_bottom=%d output_w=%d output_h=%d\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1], output_pad_right, output_pad_bottom, output_w, output_h);
74
        }
75
    }
76

77
    {
78
        ncnn::Option opt;
79
        opt.num_threads = 1;
80
        opt.use_packing_layout = true;
81
        opt.use_fp16_packed = true;
82
        opt.use_fp16_storage = true;
83
        opt.use_fp16_arithmetic = true;
84
        opt.use_bf16_storage = true;
85
        opt.use_shader_pack8 = true;
86
        opt.use_image_storage = true;
87
        opt.use_sgemm_convolution = false;
88
        opt.use_winograd_convolution = false;
89

90
        ret = test_layer_opt("Deconvolution", pd, weights, opt, a);
91
        if (ret != 0)
92
        {
93
            fprintf(stderr, "test_deconvolution failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f] output_pad_right=%d output_pad_bottom=%d output_w=%d output_h=%d\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1], output_pad_right, output_pad_bottom, output_w, output_h);
94
        }
95
    }
96

97
    return ret;
98
}
99

100
static int test_deconvolution_0()
101
{
102
    static const int kdsp[16][4] = {
103
        {1, 1, 1, 0},
104
        {1, 1, 2, 0},
105
        {2, 1, 1, 1},
106
        {2, 1, 2, -233},
107
        {3, 1, 1, 1},
108
        {3, 1, 2, 1},
109
        {3, 2, 1, 1},
110
        {4, 1, 1, -233},
111
        {4, 1, 2, -234},
112
        {4, 2, 1, -234},
113
        {5, 1, 1, 2},
114
        {5, 1, 2, 2},
115
        {5, 2, 2, 2},
116
        {7, 1, 1, 3},
117
        {7, 1, 2, 3},
118
        {7, 2, 1, -233},
119
    };
120

121
    for (int i = 0; i < 16; i++)
122
    {
123
        const int k = kdsp[i][0];
124
        const int d = kdsp[i][1];
125
        const int s = kdsp[i][2];
126
        const int p = kdsp[i][3];
127

128
        int ret = 0
129
                  || test_deconvolution(9, 7, 1, 1, k, d, s, p, 1, 0, 0, 0, 0)
130
                  || test_deconvolution(9, 7, 4, 13, k, d, s, p, 0, 1, 1, 7, 5)
131
                  || test_deconvolution(9, 7, 13, 4, k, d, s, p, 1, 1, 0, 0, 0)
132
                  || test_deconvolution(9, 7, 4, 8, k, d, s, p, 0, 0, 1, 0, 0)
133
                  || test_deconvolution(9, 7, 8, 4, k, d, s, p, 1, 0, 0, 7, 5)
134
                  || test_deconvolution(7, 7, 12, 12, k, d, s, p, 1, 0, 1, 0, 0)
135
                  || test_deconvolution(4, 5, 12, 11, k, d, s, p, 0, 0, 1, 1, 0)
136
                  || test_deconvolution(9, 7, 8, 13, k, d, s, p, 0, 2, 2, 0, 0)
137
                  || test_deconvolution(9, 7, 13, 8, k, d, s, p, 1, 2, 0, 0, 0)
138
                  || test_deconvolution(9, 7, 16, 16, k, d, s, p, 0, 0, 2, 7, 5);
139

140
        if (ret != 0)
141
            return -1;
142
    }
143

144
    return 0
145
           || test_deconvolution(7, 5, 24, 32, 4, 2, 2, 2, 1, 0, 0, 0, 0)
146
           || test_deconvolution(7, 5, 32, 24, 4, 2, 2, 2, 1, 0, 0, 0, 0)
147
           || test_deconvolution(7, 5, 28, 32, 4, 2, 2, 2, 1, 0, 0, 0, 0)
148
           || test_deconvolution(7, 5, 32, 28, 4, 2, 2, 2, 1, 0, 0, 0, 0)
149
           || test_deconvolution(7, 5, 26, 32, 4, 2, 2, 2, 1, 0, 0, 0, 0)
150
           || test_deconvolution(7, 5, 32, 26, 4, 2, 2, 2, 1, 0, 0, 0, 0);
151
}
152

153
static int test_deconvolution_dynamic(int w, int h, int c, int outch, int kernel, int dilation, int stride, int pad, int bias, int output_pad_right, int output_pad_bottom, int output_w, int output_h)
154
{
155
    ncnn::Mat a = RandomMat(w, h, c);
156

157
    if (output_w > 0 && output_h > 0 && pad != -233 && pad != -234)
158
    {
159
        pad = -233;
160
    }
161

162
    ncnn::ParamDict pd;
163
    pd.set(0, 0);
164
    pd.set(1, 0);
165
    pd.set(2, dilation);
166
    pd.set(3, stride);
167
    pd.set(4, pad);
168
    pd.set(5, bias);
169
    pd.set(6, 0);
170
    pd.set(28, 1); // dynamic weight
171

172
    int activation_type = RAND() % 7; // 0 1 2 3 4 5 6
173
    ncnn::Mat activation_params(2);
174
    activation_params[0] = (activation_type == 6) ? RandomFloat(0, 1) : RandomFloat(-1, 0); // alpha
175
    activation_params[1] = RandomFloat(0, 1);                                               // beta
176
    pd.set(9, activation_type);
177
    pd.set(10, activation_params);
178

179
    pd.set(18, output_pad_right);
180
    pd.set(19, output_pad_bottom);
181
    pd.set(20, output_w);
182
    pd.set(21, output_h);
183

184
    std::vector<ncnn::Mat> as(bias ? 3 : 2);
185
    as[0] = a;
186
    as[1] = RandomMat(kernel, kernel, outch, c);
187
    if (bias)
188
        as[2] = RandomMat(outch);
189

190
    std::vector<ncnn::Mat> weights(0);
191

192
    int ret = test_layer("Deconvolution", pd, weights, as);
193
    if (ret != 0)
194
    {
195
        fprintf(stderr, "test_deconvolution_dynamic failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f] output_pad_right=%d output_pad_bottom=%d output_w=%d output_h=%d\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1], output_pad_right, output_pad_bottom, output_w, output_h);
196
    }
197

198
    return ret;
199
}
200

201
static int test_deconvolution_1()
202
{
203
    static const int kdsp[7][4] = {
204
        {1, 1, 1, 0},
205
        {1, 1, 2, 0},
206
        {2, 1, 1, 1},
207
        {2, 1, 2, -233},
208
        {3, 1, 1, 1},
209
        {3, 1, 2, 1},
210
        {3, 2, 1, -234},
211
    };
212

213
    for (int i = 0; i < 7; i++)
214
    {
215
        const int k = kdsp[i][0];
216
        const int d = kdsp[i][1];
217
        const int s = kdsp[i][2];
218
        const int p = kdsp[i][3];
219

220
        int ret = 0
221
                  || test_deconvolution_dynamic(9, 7, 1, 1, k, d, s, p, 1, 0, 0, 0, 0)
222
                  || test_deconvolution_dynamic(9, 7, 4, 13, k, d, s, p, 0, 1, 1, 7, 5)
223
                  || test_deconvolution_dynamic(9, 7, 13, 4, k, d, s, p, 1, 1, 0, 0, 0)
224
                  || test_deconvolution_dynamic(9, 7, 4, 8, k, d, s, p, 0, 0, 1, 0, 0)
225
                  || test_deconvolution_dynamic(9, 7, 8, 4, k, d, s, p, 1, 0, 0, 7, 5)
226
                  || test_deconvolution_dynamic(7, 7, 12, 12, k, d, s, p, 1, 0, 1, 0, 0)
227
                  || test_deconvolution_dynamic(4, 5, 12, 11, k, d, s, p, 0, 0, 1, 1, 0)
228
                  || test_deconvolution_dynamic(9, 7, 8, 13, k, d, s, p, 0, 2, 2, 0, 0)
229
                  || test_deconvolution_dynamic(9, 7, 13, 8, k, d, s, p, 1, 2, 0, 0, 0)
230
                  || test_deconvolution_dynamic(9, 7, 16, 16, k, d, s, p, 0, 0, 2, 7, 5);
231

232
        if (ret != 0)
233
            return -1;
234
    }
235

236
    return 0;
237
}
238

239
int main()
240
{
241
    SRAND(7767517);
242

243
    return test_deconvolution_0() || test_deconvolution_1();
244
}
245

Использование cookies

Мы используем файлы cookie в соответствии с Политикой конфиденциальности и Политикой использования cookies.

Нажимая кнопку «Принимаю», Вы даете АО «СберТех» согласие на обработку Ваших персональных данных в целях совершенствования нашего веб-сайта и Сервиса GitVerse, а также повышения удобства их использования.

Запретить использование cookies Вы можете самостоятельно в настройках Вашего браузера.