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
|
// Copyright 2014 BVLC and contributors.
#include <vector>
#include "caffe/layer.hpp"
#include "caffe/vision_layers.hpp"
#include "caffe/util/im2col.hpp"
#include "caffe/filler.hpp"
#include "caffe/util/math_functions.hpp"
namespace caffe {
template <typename Dtype>
void ConvolutionLayer<Dtype>::SetUp(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {
Layer<Dtype>::SetUp(bottom, top);
kernel_size_ = this->layer_param_.convolution_param().kernel_size();
stride_ = this->layer_param_.convolution_param().stride();
group_ = this->layer_param_.convolution_param().group();
pad_ = this->layer_param_.convolution_param().pad();
num_ = bottom[0]->num();
channels_ = bottom[0]->channels();
height_ = bottom[0]->height();
width_ = bottom[0]->width();
// TODO: generalize to handle inputs of different shapes.
for (int bottom_id = 1; bottom_id < bottom.size(); ++bottom_id) {
CHECK_EQ(num_, bottom[bottom_id]->num()) << "Inputs must have same num.";
CHECK_EQ(channels_, bottom[bottom_id]->channels())
<< "Inputs must have same channels.";
CHECK_EQ(height_, bottom[bottom_id]->height())
<< "Inputs must have same height.";
CHECK_EQ(width_, bottom[bottom_id]->width())
<< "Inputs must have same width.";
}
num_output_ = this->layer_param_.convolution_param().num_output();
CHECK_GT(num_output_, 0);
CHECK_EQ(channels_ % group_, 0);
// The im2col result buffer would only hold one image at a time to avoid
// overly large memory usage.
int height_out = (height_ + 2 * pad_ - kernel_size_) / stride_ + 1;
int width_out = (width_ + 2 * pad_ - kernel_size_) / stride_ + 1;
col_buffer_.Reshape(
1, channels_ * kernel_size_ * kernel_size_, height_out, width_out);
// Set the parameters
CHECK_EQ(num_output_ % group_, 0)
<< "Number of output should be multiples of group.";
bias_term_ = this->layer_param_.convolution_param().bias_term();
// Figure out the dimensions for individual gemms.
M_ = num_output_ / group_;
K_ = channels_ * kernel_size_ * kernel_size_ / group_;
N_ = height_out * width_out;
for (int top_id = 0; top_id < top->size(); ++top_id) {
(*top)[top_id]->Reshape(num_, num_output_, height_out, width_out);
}
// Check if we need to set up the weights
if (this->blobs_.size() > 0) {
LOG(INFO) << "Skipping parameter initialization";
} else {
if (bias_term_) {
this->blobs_.resize(2);
} else {
this->blobs_.resize(1);
}
// Intialize the weight
this->blobs_[0].reset(new Blob<Dtype>(
num_output_, channels_ / group_, kernel_size_, kernel_size_));
// fill the weights
shared_ptr<Filler<Dtype> > weight_filler(GetFiller<Dtype>(
this->layer_param_.convolution_param().weight_filler()));
weight_filler->Fill(this->blobs_[0].get());
// If necessary, initialize and fill the bias term
if (bias_term_) {
this->blobs_[1].reset(new Blob<Dtype>(1, 1, 1, num_output_));
shared_ptr<Filler<Dtype> > bias_filler(GetFiller<Dtype>(
this->layer_param_.convolution_param().bias_filler()));
bias_filler->Fill(this->blobs_[1].get());
}
}
// Set up the all ones "bias multiplier" for adding bias using blas
if (bias_term_) {
bias_multiplier_.Reshape(1, 1, 1, N_);
caffe_set(N_, Dtype(1), bias_multiplier_.mutable_cpu_data());
}
this->param_propagate_down_.resize(this->blobs_.size(), true);
}
template <typename Dtype>
Dtype ConvolutionLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
vector<Blob<Dtype>*>* top) {
for (int i = 0; i < bottom.size(); ++i) {
const Dtype* bottom_data = bottom[i]->cpu_data();
Dtype* top_data = (*top)[i]->mutable_cpu_data();
Dtype* col_data = col_buffer_.mutable_cpu_data();
const Dtype* weight = this->blobs_[0]->cpu_data();
int weight_offset = M_ * K_;
int col_offset = K_ * N_;
int top_offset = M_ * N_;
for (int n = 0; n < num_; ++n) {
// First, im2col
im2col_cpu(bottom_data + bottom[i]->offset(n), channels_, height_,
width_, kernel_size_, pad_, stride_, col_data);
// Second, innerproduct with groups
for (int g = 0; g < group_; ++g) {
caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, M_, N_, K_,
(Dtype)1., weight + weight_offset * g, col_data + col_offset * g,
(Dtype)0., top_data + (*top)[i]->offset(n) + top_offset * g);
}
// third, add bias
if (bias_term_) {
caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasNoTrans, num_output_,
N_, 1, (Dtype)1., this->blobs_[1]->cpu_data(),
bias_multiplier_.cpu_data(),
(Dtype)1., top_data + (*top)[i]->offset(n));
}
}
}
return Dtype(0.);
}
template <typename Dtype>
void ConvolutionLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
const vector<bool>& propagate_down, vector<Blob<Dtype>*>* bottom) {
const Dtype* weight = NULL;
Dtype* weight_diff = NULL;
if (this->param_propagate_down_[0]) {
weight = this->blobs_[0]->cpu_data();
weight_diff = this->blobs_[0]->mutable_cpu_diff();
caffe_set(this->blobs_[0]->count(), Dtype(0), weight_diff);
}
Dtype* bias_diff = NULL;
if (bias_term_ && this->param_propagate_down_[1]) {
bias_diff = this->blobs_[1]->mutable_cpu_diff();
caffe_set(this->blobs_[1]->count(), Dtype(0), bias_diff);
}
const int weight_offset = M_ * K_;
const int col_offset = K_ * N_;
const int top_offset = M_ * N_;
for (int i = 0; i < top.size(); ++i) {
const Dtype* top_diff = NULL;
// Bias gradient, if necessary.
if (bias_term_ && this->param_propagate_down_[1]) {
top_diff = top[i]->cpu_diff();
for (int n = 0; n < num_; ++n) {
caffe_cpu_gemv<Dtype>(CblasNoTrans, num_output_, N_,
1., top_diff + top[0]->offset(n),
bias_multiplier_.cpu_data(), 1.,
bias_diff);
}
}
if (this->param_propagate_down_[0] || propagate_down[i]) {
if (!top_diff) {
top_diff = top[i]->cpu_diff();
}
Dtype* col_data = col_buffer_.mutable_cpu_data();
Dtype* col_diff = col_buffer_.mutable_cpu_diff();
const Dtype* bottom_data = (*bottom)[i]->cpu_data();
Dtype* bottom_diff = (*bottom)[i]->mutable_cpu_diff();
for (int n = 0; n < num_; ++n) {
// Since we saved memory in the forward pass by not storing all col
// data, we will need to recompute them.
im2col_cpu(bottom_data + (*bottom)[i]->offset(n), channels_, height_,
width_, kernel_size_, pad_, stride_, col_data);
// gradient w.r.t. weight. Note that we will accumulate diffs.
if (this->param_propagate_down_[0]) {
for (int g = 0; g < group_; ++g) {
caffe_cpu_gemm<Dtype>(CblasNoTrans, CblasTrans, M_, K_, N_,
(Dtype)1., top_diff + top[i]->offset(n) + top_offset * g,
col_data + col_offset * g, (Dtype)1.,
weight_diff + weight_offset * g);
}
}
// gradient w.r.t. bottom data, if necessary
if (propagate_down[i]) {
for (int g = 0; g < group_; ++g) {
caffe_cpu_gemm<Dtype>(CblasTrans, CblasNoTrans, K_, N_, M_,
(Dtype)1., weight + weight_offset * g,
top_diff + top[i]->offset(n) + top_offset * g,
(Dtype)0., col_diff + col_offset * g);
}
// col2im back to the data
col2im_cpu(col_diff, channels_, height_, width_, kernel_size_, pad_,
stride_, bottom_diff + (*bottom)[i]->offset(n));
}
}
}
}
}
#ifdef CPU_ONLY
STUB_GPU(ConvolutionLayer);
#endif
INSTANTIATE_CLASS(ConvolutionLayer);
} // namespace caffe
|
