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
213
214
215
216
217
218
219
220
221
222
223
|
/*
Copyright (c) 2013, Taiga Nomi
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 "tiny_dnn/util/util.h"
#include <algorithm>
namespace tiny_dnn {
enum class norm_region {
across_channels,
within_channels
};
/**
* local response normalization
*/
template<typename Activation>
class lrn_layer : public feedforward_layer<Activation> {
public:
CNN_USE_LAYER_MEMBERS;
typedef feedforward_layer<Activation> Base;
lrn_layer(const shape3d& in_shape,
serial_size_t local_size,
float_t alpha = 1.0,
float_t beta = 5.0,
norm_region region = norm_region::across_channels)
: Base({ vector_type::data }),
in_shape_(in_shape),
size_(local_size),
alpha_(alpha),
beta_(beta),
region_(region),
in_square_(in_shape_.area()) {
}
/**
* @param layer [in] the previous layer connected to this
* @param local_size [in] the number of channels(depths) to sum over
* @param in_channels [in] the number of channels of input data
* @param alpha [in] the scaling parameter (same to caffe's LRN)
* @param beta [in] the scaling parameter (same to caffe's LRN)
**/
lrn_layer(layer* prev,
serial_size_t local_size,
float_t alpha = 1.0,
float_t beta = 5.0,
norm_region region = norm_region::across_channels)
: lrn_layer(prev->out_data_shape()[0], local_size, alpha, beta, region) {}
/**
* @param in_width [in] the width of input data
* @param in_height [in] the height of input data
* @param local_size [in] the number of channels(depths) to sum over
* @param in_channels [in] the number of channels of input data
* @param alpha [in] the scaling parameter (same to caffe's LRN)
* @param beta [in] the scaling parameter (same to caffe's LRN)
**/
lrn_layer(serial_size_t in_width,
serial_size_t in_height,
serial_size_t local_size,
serial_size_t in_channels,
float_t alpha = 1.0,
float_t beta = 5.0,
norm_region region = norm_region::across_channels)
: lrn_layer(shape3d{in_width, in_height, in_channels}, local_size, alpha, beta, region){}
serial_size_t fan_in_size() const override {
return size_;
}
serial_size_t fan_out_size() const override {
return size_;
}
std::vector<shape3d> in_shape() const override {
return { in_shape_ };
}
std::vector<shape3d> out_shape() const override {
return { in_shape_, in_shape_ };
}
std::string layer_type() const override { return "lrn"; }
void forward_propagation(const std::vector<tensor_t*>& in_data,
std::vector<tensor_t*>& out_data) override {
// @todo revise the parallelism strategy
for (size_t sample = 0, sample_count = in_data[0]->size(); sample < sample_count; ++sample) {
vec_t& in = (*in_data[0])[sample];
vec_t& out = (*out_data[0])[sample];
vec_t& a = (*out_data[1])[sample];
if (region_ == norm_region::across_channels) {
forward_across(in, a);
}
else {
forward_within(in, a);
}
for_i(parallelize_, out.size(), [&](int i) {
out[i] = h_.f(a, i);
});
}
}
void back_propagation(const std::vector<tensor_t*>& in_data,
const std::vector<tensor_t*>& out_data,
std::vector<tensor_t*>& out_grad,
std::vector<tensor_t*>& in_grad) override {
CNN_UNREFERENCED_PARAMETER(in_data);
CNN_UNREFERENCED_PARAMETER(out_data);
CNN_UNREFERENCED_PARAMETER(out_grad);
CNN_UNREFERENCED_PARAMETER(in_grad);
throw nn_error("not implemented");
}
template <class Archive>
static void load_and_construct(Archive & ar, cereal::construct<lrn_layer> & construct) {
shape3d in_shape;
serial_size_t size;
float_t alpha, beta;
norm_region region;
ar(cereal::make_nvp("in_shape", in_shape),
cereal::make_nvp("size", size),
cereal::make_nvp("alpha", alpha),
cereal::make_nvp("beta", beta),
cereal::make_nvp("region", region));
construct(in_shape, size, alpha, beta, region);
}
template <class Archive>
void serialize(Archive & ar) {
layer::serialize_prolog(ar);
ar(cereal::make_nvp("in_shape", in_shape_),
cereal::make_nvp("size", size_),
cereal::make_nvp("alpha", alpha_),
cereal::make_nvp("beta", beta_),
cereal::make_nvp("region", region_));
}
private:
void forward_across(const vec_t& in, vec_t& out) {
std::fill(in_square_.begin(), in_square_.end(), float_t(0));
for (serial_size_t i = 0; i < size_ / 2; i++) {
serial_size_t idx = in_shape_.get_index(0, 0, i);
add_square_sum(&in[idx], in_shape_.area(), &in_square_[0]);
}
serial_size_t head = size_ / 2;
long tail = static_cast<long>(head) - static_cast<long>(size_);
serial_size_t channels = in_shape_.depth_;
const serial_size_t wxh = in_shape_.area();
const float_t alpha_div_size = alpha_ / size_;
for (serial_size_t i = 0; i < channels; i++, head++, tail++) {
if (head < channels)
add_square_sum(&in[in_shape_.get_index(0, 0, head)], wxh, &in_square_[0]);
if (tail >= 0)
sub_square_sum(&in[in_shape_.get_index(0, 0, tail)], wxh, &in_square_[0]);
float_t *dst = &out[in_shape_.get_index(0, 0, i)];
const float_t *src = &in[in_shape_.get_index(0, 0, i)];
for (serial_size_t j = 0; j < wxh; j++)
dst[j] = src[j] * std::pow(float_t(1) + alpha_div_size * in_square_[j], -beta_);
}
}
void forward_within(const vec_t& in, vec_t& out) {
CNN_UNREFERENCED_PARAMETER(in);
CNN_UNREFERENCED_PARAMETER(out);
throw nn_error("not implemented");
}
void add_square_sum(const float_t *src, serial_size_t size, float_t *dst) {
for (serial_size_t i = 0; i < size; i++)
dst[i] += src[i] * src[i];
}
void sub_square_sum(const float_t *src, serial_size_t size, float_t *dst) {
for (serial_size_t i = 0; i < size; i++)
dst[i] -= src[i] * src[i];
}
shape3d in_shape_;
serial_size_t size_;
float_t alpha_, beta_;
norm_region region_;
vec_t in_square_;
};
} // namespace tiny_dnn
|
