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
|
/*
Copyright (c) 2016, Taiga Nomi, Edgar Riba
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of the <organization> nor the
names of its contributors may be used to endorse or promote products
derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY
EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY
DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#pragma once
#include "params.h"
namespace tiny_dnn {
namespace core {
struct conv_layer_worker_specific_storage {
std::vector<const vec_t*> prev_out_padded_;
std::vector<vec_t> prev_out_buf_;
std::vector<vec_t> prev_delta_padded_;
};
struct connection_table {
connection_table() : rows_(0), cols_(0) {}
connection_table(const bool *ar, serial_size_t rows, serial_size_t cols)
: connected_(rows * cols), rows_(rows), cols_(cols) {
std::copy(ar, ar + rows * cols, connected_.begin());
}
connection_table(serial_size_t ngroups, serial_size_t rows, serial_size_t cols)
: connected_(rows * cols, false), rows_(rows), cols_(cols) {
if (rows % ngroups || cols % ngroups) {
throw nn_error("invalid group size");
}
serial_size_t row_group = rows / ngroups;
serial_size_t col_group = cols / ngroups;
serial_size_t idx = 0;
for (serial_size_t g = 0; g < ngroups; g++) {
for (serial_size_t r = 0; r < row_group; r++) {
for (serial_size_t c = 0; c < col_group; c++) {
idx = (r + g * row_group) * cols_ + c + g * col_group;
connected_[idx] = true;
}
}
}
}
bool is_connected(serial_size_t x, serial_size_t y) const {
return is_empty() ? true : connected_[y * cols_ + x];
}
bool is_empty() const {
return rows_ == 0 && cols_ == 0;
}
template <typename Archive>
void serialize(Archive & ar) {
ar(cereal::make_nvp("rows", rows_), cereal::make_nvp("cols", cols_));
if (is_empty()) {
ar(cereal::make_nvp("connection", std::string("all")));
}
else {
ar(cereal::make_nvp("connection", connected_));
}
}
std::deque<bool> connected_;
serial_size_t rows_;
serial_size_t cols_;
};
class conv_params : public Params {
public:
connection_table tbl;
index3d<serial_size_t> in;
index3d<serial_size_t> in_padded;
index3d<serial_size_t> out;
index3d<serial_size_t> weight;
bool has_bias;
padding pad_type;
serial_size_t w_stride;
serial_size_t h_stride;
friend std::ostream& operator<<(std::ostream &o,
const core::conv_params& param) {
o << "in: " << param.in << "\n";
o << "out: " << param.out << "\n";
o << "in_padded: " << param.in_padded << "\n";
o << "weight: " << param.weight << "\n";
o << "has_bias: " << param.has_bias << "\n";
o << "w_stride: " << param.w_stride << "\n";
o << "h_stride: " << param.h_stride << "\n";
return o;
}
};
inline conv_params Params::conv() const {
return *(static_cast<const conv_params*>(this));
}
class Conv2dPadding {
public:
Conv2dPadding() {}
Conv2dPadding(const conv_params& params) : params_(params) {}
/* Applies padding to an input tensor given the convolution parameters
*
* @param in The input tensor
* @param out The output tensor with padding applied
*/
void copy_and_pad_input(const tensor_t& in, tensor_t& out) {
if (params_.pad_type == padding::valid) {
return;
}
tensor_t buf(in.size());
for_i(true, buf.size(), [&](int sample) {
// alloc temporary buffer.
buf[sample].resize(params_.in_padded.size());
// make padded version in order to avoid corner-case in fprop/bprop
for (serial_size_t c = 0; c < params_.in.depth_; c++) {
float_t* pimg = &buf[sample][params_.in_padded.get_index(
params_.weight.width_ / 2,
params_.weight.height_ / 2, c)];
const float_t* pin = &in[sample][params_.in.get_index(0, 0, c)];
for (serial_size_t y = 0; y < params_.in.height_; y++) {
std::copy(pin, pin + params_.in.width_, pimg);
pin += params_.in.width_;
pimg += params_.in_padded.width_;
}
}
});
// shrink buffer to output
out = buf;
}
/* Applies unpadding to an input tensor given the convolution parameters
*
* @param in The input tensor
* @param out The output tensor with unpadding applied
*/
void copy_and_unpad_delta(const tensor_t& delta, tensor_t& delta_unpadded) {
if (params_.pad_type == padding::valid) {
return;
}
tensor_t buf(delta.size());
for_i(true, buf.size(), [&](int sample) {
// alloc temporary buffer.
buf[sample].resize(params_.in.size());
for (serial_size_t c = 0; c < params_.in.depth_; c++) {
const float_t *pin =
&delta[sample][params_.in_padded.get_index(
params_.weight.width_ / 2,
params_.weight.height_ / 2, c)];
float_t *pdst = &buf[sample][params_.in.get_index(0, 0, c)];
for (serial_size_t y = 0; y < params_.in.height_; y++) {
std::copy(pin, pin + params_.in.width_, pdst);
pdst += params_.in.width_;
pin += params_.in_padded.width_;
}
}
});
// shrink buffer to output
delta_unpadded = buf;
}
private:
conv_params params_;
};
} // namespace core
} // namespace tiny_dnn
|
