Hugging Face's logo

camenduru
/
ncnn

Model card Files
xet
Community
ncnn
File size: 6,519 Bytes
be903e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
 
// Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.

#include "deconvolution.h"

#include "fused_activation.h"

namespace ncnn {

Deconvolution::Deconvolution()
{
    one_blob_only = true;
    support_inplace = false;
}

int Deconvolution::load_param(const ParamDict& pd)
{
    num_output = pd.get(0, 0);
    kernel_w = pd.get(1, 0);
    kernel_h = pd.get(11, kernel_w);
    dilation_w = pd.get(2, 1);
    dilation_h = pd.get(12, dilation_w);
    stride_w = pd.get(3, 1);
    stride_h = pd.get(13, stride_w);
    pad_left = pd.get(4, 0);
    pad_right = pd.get(15, pad_left);
    pad_top = pd.get(14, pad_left);
    pad_bottom = pd.get(16, pad_top);
    output_pad_right = pd.get(18, 0);
    output_pad_bottom = pd.get(19, output_pad_right);
    output_w = pd.get(20, 0);
    output_h = pd.get(21, output_w);
    bias_term = pd.get(5, 0);
    weight_data_size = pd.get(6, 0);
    activation_type = pd.get(9, 0);
    activation_params = pd.get(10, Mat());

    return 0;
}

int Deconvolution::load_model(const ModelBin& mb)
{
    weight_data = mb.load(weight_data_size, 0);
    if (weight_data.empty())
        return -100;

    if (bias_term)
    {
        bias_data = mb.load(num_output, 1);
        if (bias_data.empty())
            return -100;
    }

    return 0;
}

static int deconvolution(const Mat& bottom_blob, Mat& top_blob, const Mat& weight_data, const Mat& bias_data, int kernel_w, int kernel_h, int stride_w, int stride_h, int dilation_w, int dilation_h, int activation_type, const Mat& activation_params, const Option& opt)
{
    const int outw = top_blob.w;
    const int outch = top_blob.c;

    const int maxk = kernel_w * kernel_h;

    // kernel offsets
    std::vector<int> _space_ofs(maxk);
    int* space_ofs = &_space_ofs[0];
    {
        int p1 = 0;
        int p2 = 0;
        int gap = outw * dilation_h - kernel_w * dilation_w;
        for (int i = 0; i < kernel_h; i++)
        {
            for (int j = 0; j < kernel_w; j++)
            {
                space_ofs[p1] = p2;
                p1++;
                p2 += dilation_w;
            }
            p2 += gap;
        }
    }

    #pragma omp parallel for num_threads(opt.num_threads)
    for (int p = 0; p < outch; p++)
    {
        Mat out = top_blob.channel(p);

        const float bias = bias_data.empty() ? 0.f : bias_data[p];

        out.fill(bias);

        // shadowed variable for less openmp task args
        const int w = bottom_blob.w;
        const int h = bottom_blob.h;
        const int inch = bottom_blob.c;
        const int outw = top_blob.w;
        const int outh = top_blob.h;

        for (int i = 0; i < h; i++)
        {
            for (int j = 0; j < w; j++)
            {
                float* outptr = out.row(i * stride_h) + j * stride_w;

                const float* kptr = (const float*)weight_data + maxk * inch * p;

                for (int q = 0; q < inch; q++)
                {
                    const float val = bottom_blob.channel(q).row(i)[j];

                    for (int k = 0; k < maxk; k++)
                    {
                        float w = kptr[k];
                        outptr[space_ofs[k]] += val * w;
                    }

                    kptr += maxk;
                }
            }
        }

        {
            float* outptr = out;
            int size = outw * outh;

            for (int i = 0; i < size; i++)
            {
                outptr[i] = activation_ss(outptr[i], activation_type, activation_params);
            }
        }
    }

    return 0;
}

int Deconvolution::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) const
{
    int w = bottom_blob.w;
    int h = bottom_blob.h;
    size_t elemsize = bottom_blob.elemsize;

    const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1;
    const int kernel_extent_h = dilation_h * (kernel_h - 1) + 1;

    int outw = (w - 1) * stride_w + kernel_extent_w + output_pad_right;
    int outh = (h - 1) * stride_h + kernel_extent_h + output_pad_bottom;

    Mat top_blob_bordered;
    if (pad_left > 0 || pad_right > 0 || pad_top > 0 || pad_bottom > 0 || (output_w > 0 && output_h > 0))
    {
        top_blob_bordered.create(outw, outh, num_output, elemsize, opt.workspace_allocator);
    }
    else
    {
        top_blob_bordered = top_blob;
        top_blob_bordered.create(outw, outh, num_output, elemsize, opt.blob_allocator);
    }
    if (top_blob_bordered.empty())
        return -100;

    int ret = deconvolution(bottom_blob, top_blob_bordered, weight_data, bias_data, kernel_w, kernel_h, stride_w, stride_h, dilation_w, dilation_h, activation_type, activation_params, opt);
    if (ret != 0)
        return ret;

    cut_padding(top_blob_bordered, top_blob, opt);
    if (top_blob.empty())
        return -100;

    return 0;
}

void Deconvolution::cut_padding(const Mat& top_blob_bordered, Mat& top_blob, const Option& opt) const
{
    if (pad_left > 0 || pad_right > 0 || pad_top > 0 || pad_bottom > 0)
    {
        copy_cut_border(top_blob_bordered, top_blob, pad_top, pad_bottom, pad_left, pad_right, opt);
    }
    else if (output_w > 0 && output_h > 0)
    {
        int wcut = top_blob_bordered.w - output_w;
        int hcut = top_blob_bordered.h - output_h;

        if (pad_left == -233 || pad_right == -233 || pad_top == -233 || pad_bottom == -233)
        {
            // onnx padding=SAME_UPPER
            copy_cut_border(top_blob_bordered, top_blob, hcut / 2, hcut - hcut / 2, wcut / 2, wcut - wcut / 2, opt);
        }
        else if (pad_left == -234 || pad_right == -234 || pad_top == -234 || pad_bottom == -234)
        {
            // onnx padding=SAME_LOWER
            copy_cut_border(top_blob_bordered, top_blob, hcut - hcut / 2, hcut / 2, wcut - wcut / 2, wcut / 2, opt);
        }
    }
    else
    {
        top_blob = top_blob_bordered;
    }
}

} // namespace ncnn