caffe源码追踪-net

#ifndef CAFFE_NET_HPP_
#define CAFFE_NET_HPP_

#include <map>
#include <set>
#include <string>
#include <utility>
#include <vector>

#include "caffe/blob.hpp"
#include "caffe/common.hpp"
#include "caffe/layer.hpp"
#include "caffe/proto/caffe.pb.h"

namespace caffe {
template <typename Dtype>
class Net {
public:
explicit Net(const NetParameter& param, const Net* root_net = NULL);//构造函数
explicit Net(const string& param_file, Phase phase,
const int level = 0, const vector<string>* stages = NULL,
const Net* root_net = NULL);
virtual ~Net() {}//虚析构函数
void Init(const NetParameter& param);//用网络参数初始化网络
const vector<Blob<Dtype>*>& Forward(Dtype* loss = NULL);//前向传播
const vector<Blob<Dtype>*>& ForwardPrefilled(Dtype* loss = NULL) {//不建议使用,用上面代替
LOG_EVERY_N(WARNING, 1000) << "DEPRECATED: ForwardPrefilled() "
<< "will be removed in a future version. Use Forward().";
return Forward(loss);
}
Dtype ForwardFromTo(int start, int end);
Dtype ForwardFrom(int start);
Dtype ForwardTo(int end);
const vector<Blob<Dtype>*>& Forward(const vector<Blob<Dtype>* > & bottom,
Dtype* loss = NULL);//不建议使用,用Forward()代替
void ClearParamDiffs();//清空所有梯度,backward之前必要的一步
void Backward();//接下来都是反向传播
void BackwardFromTo(int start, int end);
void BackwardFrom(int start);
void BackwardTo(int end);
void Reshape();//从输入到输出层层塑形
Dtype ForwardBackward() {
Dtype loss;
Forward(&loss);
Backward();
return loss;
}
void Update();//更新一次网络权重
void ShareWeights();//初始化会调用该函数,用了共享权重
void ShareTrainedLayersWith(const Net* other);//共享其他网络的一些层
void CopyTrainedLayersFrom(const NetParameter& param);
void CopyTrainedLayersFrom(const string trained_filename);
void CopyTrainedLayersFromBinaryProto(const string trained_filename);
void CopyTrainedLayersFromHDF5(const string trained_filename);
void ToProto(NetParameter* param, bool write_diff = false) const;//序列化网络为protobuf
void ToHDF5(const string& filename, bool write_diff = false) const;//将网络转换到HDF5文件
inline const string& name() const { return name_; }//获取网络名字
inline const vector<string>& layer_names() const { return layer_names_; }//获取层的名字
inline const vector<string>& blob_names() const { return blob_names_; }//获取blob名字
inline const vector<shared_ptr<Blob<Dtype> > >& blobs() const {//获取指向blob数据的指针
return blobs_;
}
inline const vector<shared_ptr<Layer<Dtype> > >& layers() const {//获取指向layer数据的指针
return layers_;
}
inline Phase phase() const { return phase_; }//获取阶段
inline const vector<vector<Blob<Dtype>*> >& bottom_vecs() const {//获取每一层输入向量,通常不需要,除非是做一些训练前的检查
return bottom_vecs_;
}
inline const vector<vector<Blob<Dtype>*> >& top_vecs() const {//获取每一层输出向量,通常不需要,除非是做一些训练前的检查
return top_vecs_;
}
inline const vector<int> & top_ids(int i) const {//获取第i层的输出blob id
CHECK_GE(i, 0) << "Invalid layer id";
CHECK_LT(i, top_id_vecs_.size()) << "Invalid layer id";
return top_id_vecs_[i];
}
inline const vector<int> & bottom_ids(int i) const {//获取第i层的输入blob id
CHECK_GE(i, 0) << "Invalid layer id";
CHECK_LT(i, bottom_id_vecs_.size()) << "Invalid layer id";
return bottom_id_vecs_[i];
}
inline const vector<vector<bool> >& bottom_need_backward() const {//输入是否需要反向传播
return bottom_need_backward_;
}
inline const vector<Dtype>& blob_loss_weights() const {//获取loss_weights
return blob_loss_weights_;
}
inline const vector<bool>& layer_need_backward() const {//该层是否需要反向传播
return layer_need_backward_;
}
inline const vector<shared_ptr<Blob<Dtype> > >& params() const {//获取参数
return params_;
}
inline const vector<Blob<Dtype>*>& learnable_params() const {//获取可学习参数
return learnable_params_;
}
inline const vector<float>& params_lr() const { return params_lr_; }//获取学习率
inline const vector<bool>& has_params_lr() const { return has_params_lr_; }//是否有学习率
inline const vector<float>& params_weight_decay() const {//获取可学习参数decay
return params_weight_decay_;
}
inline const vector<bool>& has_params_decay() const {
return has_params_decay_;
}
const map<string, int>& param_names_index() const {
return param_names_index_;
}
inline const vector<int>& param_owners() const { return param_owners_; }
inline const vector<string>& param_display_names() const {
return param_display_names_;
}
// 接下来就是一些获取输入输出大小
inline int num_inputs() const { return net_input_blobs_.size(); }
inline int num_outputs() const { return net_output_blobs_.size(); }
inline const vector<Blob<Dtype>*>& input_blobs() const {
return net_input_blobs_;
}
inline const vector<Blob<Dtype>*>& output_blobs() const {
return net_output_blobs_;
}
inline const vector<int>& input_blob_indices() const {
return net_input_blob_indices_;
}
inline const vector<int>& output_blob_indices() const {
return net_output_blob_indices_;
}
bool has_blob(const string& blob_name) const;
const shared_ptr<Blob<Dtype> > blob_by_name(const string& blob_name) const;
bool has_layer(const string& layer_name) const;
const shared_ptr<Layer<Dtype> > layer_by_name(const string& layer_name) const;
void set_debug_info(const bool value) { debug_info_ = value; }
static void FilterNet(const NetParameter& param,//移走一些现阶段不需要的层,具体见cpp文件
NetParameter* param_filtered);
static bool StateMeetsRule(const NetState& state, const NetStateRule& rule,
const string& layer_name);//判断state是否满足rule

protected:
void AppendTop(const NetParameter& param, const int layer_id,//为网络添加新的输出
const int top_id, set<string>* available_blobs,
map<string, int>* blob_name_to_idx);
int AppendBottom(const NetParameter& param, const int layer_id,//为网络添加新的输入
const int bottom_id, set<string>* available_blobs,
map<string, int>* blob_name_to_idx);
void AppendParam(const NetParameter& param, const int layer_id,//为网络添加新的参数
const int param_id);
void ForwardDebugInfo(const int layer_id);//是否打印调试信息在Forward过程中
void BackwardDebugInfo(const int layer_id);//是否打印调试信息在Backward过程中
void UpdateDebugInfo(const int param_id);//是否打印调试信息在 Update过程中
string name_;//网络的名字
Phase phase_;//阶段
vector<shared_ptr<Layer<Dtype> > > layers_;//所有层
vector<string> layer_names_;//每个层名字
map<string, int> layer_names_index_;//层的名字及其索引
vector<bool> layer_need_backward_;//每个层是否需要backward
//接下来是一些存储层之间的中间结果的容器
vector<shared_ptr<Blob<Dtype> > > blobs_;//
vector<string> blob_names_;//blob名字
map<string, int> blob_names_index_;
vector<bool> blob_need_backward_;//blob是否需要backward
vector<vector<Blob<Dtype>*> > bottom_vecs_;//实际数据存储在blobs_里面,这里只是装的指针:假如有n个层,则bottom_vecs_装的是n个指针,每个指针指向具体输入数据
vector<vector<int> > bottom_id_vecs_;
vector<vector<bool> > bottom_need_backward_;//存放各层各输入是否需要backward
vector<vector<Blob<Dtype>*> > top_vecs_;
vector<vector<int> > top_id_vecs_;
vector<Dtype> blob_loss_weights_;
vector<vector<int> > param_id_vecs_;
vector<int> param_owners_;
vector<string> param_display_names_;
vector<pair<int, int> > param_layer_indices_;
map<string, int> param_names_index_;
vector<int> net_input_blob_indices_;
vector<int> net_output_blob_indices_;
vector<Blob<Dtype>*> net_input_blobs_;
vector<Blob<Dtype>*> net_output_blobs_;
vector<shared_ptr<Blob<Dtype> > > params_;
vector<Blob<Dtype>*> learnable_params_;
vector<int> learnable_param_ids_;
vector<float> params_lr_;
vector<bool> has_params_lr_;
vector<float> params_weight_decay_;
vector<bool> has_params_decay_;
size_t memory_used_;
bool debug_info_;
const Net* const root_net_;
DISABLE_COPY_AND_ASSIGN(Net);
};
}  // namespace caffe

#endif  // CAFFE_NET_HPP_

#include <algorithm>
#include <map>
#include <set>
#include <string>
#include <utility>
#include <vector>

#include "hdf5.h"

#include "caffe/common.hpp"
#include "caffe/layer.hpp"
#include "caffe/net.hpp"
#include "caffe/parallel.hpp"
#include "caffe/proto/caffe.pb.h"
#include "caffe/util/hdf5.hpp"
#include "caffe/util/insert_splits.hpp"
#include "caffe/util/math_functions.hpp"
#include "caffe/util/upgrade_proto.hpp"

#include "caffe/test/test_caffe_main.hpp"

namespace caffe {
template <typename Dtype>
Net<Dtype>::Net(const NetParameter& param, const Net* root_net)//root_net与多GPU并行训练有关,暂时忽略,接受NetParameter输入初始化网络
: root_net_(root_net) {
Init(param);
}
template <typename Dtype>
Net<Dtype>::Net(const string& param_file, Phase phase,//接受prototxt文件作为输入
const int level, const vector<string>* stages,
const Net* root_net)
: root_net_(root_net) {
NetParameter param;
ReadNetParamsFromTextFileOrDie(param_file, ¶m);
param.mutable_state()->set_phase(phase);
if (stages != NULL) {
for (int i = 0; i < stages->size(); i++) {
param.mutable_state()->add_stage((*stages)[i]);
}
}
param.mutable_state()->set_level(level);
Init(param);
}
template <typename Dtype>
void Net<Dtype>::Init(const NetParameter& in_param) {
CHECK(Caffe::root_solver() || root_net_)
<< "root_net_ needs to be set for all non-root solvers";
phase_ = in_param.state().phase();
NetParameter filtered_param;
FilterNet(in_param, &filtered_param);//过滤网络,filtered_param即为过滤后的
LOG_IF(INFO, Caffe::root_solver())
<< "Initializing net from parameters: " << std::endl
<< filtered_param.DebugString();
NetParameter param;
InsertSplits(filtered_param, ¶m);//对于多个输入一个输出的blob需要加入分裂层,为了在反向传播时将梯度累加,函数声明在caffe/util/insert_splits.hpp文件
name_ = param.name();//网络名字
map<string, int> blob_name_to_idx;//blob名字与id,整个网络的输入输出blob名字及其id,不重复(in-place计算可以)
set<string> available_blobs;//可用的blob,暂时还没有用到过的,存放的是blob的名字,用了则删除该blob名字
memory_used_ = 0;//内存使用初始化为0
bottom_vecs_.resize(param.layer_size());//假如有n个层,则bottom_vecs_装的是n个指针,每个指针指向具体输入数据,所以大小与层数一致
top_vecs_.resize(param.layer_size());//同理
bottom_id_vecs_.resize(param.layer_size());//同理
param_id_vecs_.resize(param.layer_size());//同理
top_id_vecs_.resize(param.layer_size());//同理
bottom_need_backward_.resize(param.layer_size());//同理
for (int layer_id = 0; layer_id < param.layer_size(); ++layer_id) {//逐层开始
bool share_from_root = !Caffe::root_solver()
&& root_net_->layers_[layer_id]->ShareInParallel();
if (!param.layer(layer_id).has_phase()) {
param.mutable_layer(layer_id)->set_phase(phase_);
}
const LayerParameter& layer_param = param.layer(layer_id);//layer_param为当前层
if (layer_param.propagate_down_size() > 0) {
CHECK_EQ(layer_param.propagate_down_size(),//要么不传递,为0;要么传递,与输入blob大小相等
layer_param.bottom_size())
<< "propagate_down param must be specified "
<< "either 0 or bottom_size times ";
}
if (share_from_root) {
LOG(INFO) << "Sharing layer " << layer_param.name() << " from root net";
layers_.push_back(root_net_->layers_[layer_id]);
layers_[layer_id]->SetShared(true);
} else {
layers_.push_back(LayerRegistry<Dtype>::CreateLayer(layer_param));//加层
}
layer_names_.push_back(layer_param.name());//层名字压入layer_names_
LOG_IF(INFO, Caffe::root_solver())
<< "Creating Layer " << layer_param.name();
bool need_backward = false;
for (int bottom_id = 0; bottom_id < layer_param.bottom_size();
++bottom_id) {//添加属于该层的每个输入blob
const int blob_id = AppendBottom(param, layer_id, bottom_id,
&available_blobs, &blob_name_to_idx);
need_backward |= blob_need_backward_[blob_id];//只要一个输入需要back,则整个层设置为需要back
}
int num_top = layer_param.top_size();//该层的输出blob
for (int top_id = 0; top_id < num_top; ++top_id) {//添加属于该层的每个输出blob,最开始的data层,没有输入,因此实际先AppendTop
AppendTop(param, layer_id, top_id, &available_blobs, &blob_name_to_idx);
if (layer_param.type() == "Input") {//如果层类型是Input,就将该层作为网络输入
const int blob_id = blobs_.size() - 1;
net_input_blob_indices_.push_back(blob_id);//将该层所有输出blob_id压入net_input_blob_indices_
net_input_blobs_.push_back(blobs_[blob_id].get());//将该层所有指向输出数据的指针放进net_input_blobs_
}
}
Layer<Dtype>* layer = layers_[layer_id].get();//layer指向当前层
if (layer->AutoTopBlobs()) {//如果该层设置的AutoTopBlobs()为true,则如果层参数给的输出blob个数少于设置(比如ExactNumTopBlobs(),MinTopBlobs()之类)的个数,则开辟空间
const int needed_num_top =
std::max(layer->MinTopBlobs(), layer->ExactNumTopBlobs());
for (; num_top < needed_num_top; ++num_top) {
AppendTop(param, layer_id, num_top, NULL, NULL);//添加匿名Blob,不更新available_blobs和blob_name_to_idx,因为我们不想让这些未知层被其他层所用
}
}
if (share_from_root) {
// Set up size of top blobs using root_net_
const vector<Blob<Dtype>*>& base_top = root_net_->top_vecs_[layer_id];
const vector<Blob<Dtype>*>& this_top = this->top_vecs_[layer_id];
for (int top_id = 0; top_id < base_top.size(); ++top_id) {
this_top[top_id]->ReshapeLike(*base_top[top_id]);
LOG(INFO) << "Created top blob " << top_id << " (shape: "
<< this_top[top_id]->shape_string() <<  ") for shared layer "
<< layer_param.name();
}
} else {
layers_[layer_id]->SetUp(bottom_vecs_[layer_id], top_vecs_[layer_id]);//层连接起来了,进入各层内部的SetUp函数
}
LOG_IF(INFO, Caffe::root_solver())
<< "Setting up " << layer_names_[layer_id];
for (int top_id = 0; top_id < top_vecs_[layer_id].size(); ++top_id) {
if (blob_loss_weights_.size() <= top_id_vecs_[layer_id][top_id]) {//为blob_loss_weights_开辟空间,并初始化为0
blob_loss_weights_.resize(top_id_vecs_[layer_id][top_id] + 1, Dtype(0));
}
blob_loss_weights_[top_id_vecs_[layer_id][top_id]] = layer->loss(top_id);//将每层loss值放进blob_loss_weights_里面,每个blob有个loss值
LOG_IF(INFO, Caffe::root_solver())
<< "Top shape: " << top_vecs_[layer_id][top_id]->shape_string();
if (layer->loss(top_id)) {
LOG_IF(INFO, Caffe::root_solver())
<< "    with loss weight " << layer->loss(top_id);
}
memory_used_ += top_vecs_[layer_id][top_id]->count();//使用的内存为所有输出blob数据的总和
}
LOG_IF(INFO, Caffe::root_solver())
<< "Memory required for data: " << memory_used_ * sizeof(Dtype);
const int param_size = layer_param.param_size();//该层ParamSpec类型参数的个数(该参数里面常用的是lr_mult和decay_mult)
const int num_param_blobs = layers_[layer_id]->blobs().size();//该层可学习参数种类(权重与偏置等),param_size<=num_param_blobs
CHECK_LE(param_size, num_param_blobs)
<< "Too many params specified for layer " << layer_param.name();
ParamSpec default_param_spec;
for (int param_id = 0; param_id < num_param_blobs; ++param_id) {
const ParamSpec* param_spec = (param_id < param_size) ?
&layer_param.param(param_id) : &default_param_spec;
const bool param_need_backward = param_spec->lr_mult() != 0;//学习率不为0,需要反向传播,反之,不需要
need_backward |= param_need_backward;//只要有一种参数需要反向传播,则该层需要backward
layers_[layer_id]->set_param_propagate_down(param_id,
param_need_backward);//设置该层的该种参数是否需要反向传播
}
for (int param_id = 0; param_id < num_param_blobs; ++param_id) {
AppendParam(param, layer_id, param_id);//对该层的每种参数进行添加
}
layer_need_backward_.push_back(need_backward);//最后将该层的是否需要backward信息进行保存
if (need_backward) {//如果需要反向传播,
for (int top_id = 0; top_id < top_id_vecs_[layer_id].size(); ++top_id) {
blob_need_backward_[top_id_vecs_[layer_id][top_id]] = true;//更新 blob_need_backward_(之前压入默认值false)
}
}
}
set<string> blobs_under_loss;//记录需要反向传播的输入Blob名字
set<string> blobs_skip_backp;//记录不需要反向传播的输入Blob名字
for (int layer_id = layers_.size() - 1; layer_id >= 0; --layer_id) {//逐层进行
bool layer_contributes_loss = false;//判断该层对损失是否有贡献,初始化为没有贡献
bool layer_skip_propagate_down = true;//判断该层是否需要反向传播,初始化为不需要反向
for (int top_id = 0; top_id < top_vecs_[layer_id].size(); ++top_id) {//对该层的每个输出
const string& blob_name = blob_names_[top_id_vecs_[layer_id][top_id]];//获取该输出的名字
if (layers_[layer_id]->loss(top_id) ||
(blobs_under_loss.find(blob_name) != blobs_under_loss.end())) {//如果该层存在损失或者该输出存在损失,则将该层对损失设为有贡献
layer_contributes_loss = true;
}
if (blobs_skip_backp.find(blob_name) == blobs_skip_backp.end()) {//如果该输出需要反向传播则将该层设为需要反向传播
layer_skip_propagate_down = false;
}
if (layer_contributes_loss && !layer_skip_propagate_down)//如果该层对损失有贡献且需要bp则退出内层循环
break;
}
if (layer_need_backward_[layer_id] && layer_skip_propagate_down) {//一旦设置了该层不传递梯度(layer_skip_propagate_down),则将该层及其所有输入设为不需要bp
layer_need_backward_[layer_id] = false;
for (int bottom_id = 0; bottom_id < bottom_vecs_[layer_id].size();
++bottom_id) {
bottom_need_backward_[layer_id][bottom_id] = false;
}
}
if (!layer_contributes_loss) { layer_need_backward_[layer_id] = false; }//如果该层对损失没有贡献,同样将该层设为不需要bp
if (Caffe::root_solver()) {
if (layer_need_backward_[layer_id]) {
LOG(INFO) << layer_names_[layer_id] << " needs backward computation.";
} else {
LOG(INFO) << layer_names_[layer_id]
<< " does not need backward computation.";
}
}
for (int bottom_id = 0; bottom_id < bottom_vecs_[layer_id].size();
++bottom_id) {
if (layer_contributes_loss) {
const string& blob_name =
blob_names_[bottom_id_vecs_[layer_id][bottom_id]];
blobs_under_loss.insert(blob_name);//将需要反向传播的输入Blob名字记录到blobs_under_loss 里面
} else {
bottom_need_backward_[layer_id][bottom_id] = false;
}
if (!bottom_need_backward_[layer_id][bottom_id]) {
const string& blob_name =
blob_names_[bottom_id_vecs_[layer_id][bottom_id]];
blobs_skip_backp.insert(blob_name);//将需要反向传播的输入Blob名字记录到blobs_skip_backp里面
}
}
}
if (param.force_backward()) {//如果设置了强制反向传播,则更新相应标志
for (int layer_id = 0; layer_id < layers_.size(); ++layer_id) {
layer_need_backward_[layer_id] = true;
for (int bottom_id = 0;
bottom_id < bottom_need_backward_[layer_id].size(); ++bottom_id) {
bottom_need_backward_[layer_id][bottom_id] =
bottom_need_backward_[layer_id][bottom_id] ||
layers_[layer_id]->AllowForceBackward(bottom_id);
blob_need_backward_[bottom_id_vecs_[layer_id][bottom_id]] =
blob_need_backward_[bottom_id_vecs_[layer_id][bottom_id]] ||
bottom_need_backward_[layer_id][bottom_id];
}
for (int param_id = 0; param_id < layers_[layer_id]->blobs().size();
++param_id) {
layers_[layer_id]->set_param_propagate_down(param_id, true);
}
}
}
//最后,available_blobs所有未使用blob,就是那些没有作为其他层输入的blob,被视为网络输出blob
for (set<string>::iterator it = available_blobs.begin();
it != available_blobs.end(); ++it) {
LOG_IF(INFO, Caffe::root_solver())
<< "This network produces output " << *it;
net_output_blobs_.push_back(blobs_[blob_name_to_idx[*it]].get());
net_output_blob_indices_.push_back(blob_name_to_idx[*it]);
}
for (size_t blob_id = 0; blob_id < blob_names_.size(); ++blob_id) {
blob_names_index_[blob_names_[blob_id]] = blob_id;
}
for (size_t layer_id = 0; layer_id < layer_names_.size(); ++layer_id) {
layer_names_index_[layer_names_[layer_id]] = layer_id;
}
ShareWeights();//共享权重
debug_info_ = param.debug_info();
LOG_IF(INFO, Caffe::root_solver()) << "Network initialization done.";
}

template <typename Dtype>
void Net<Dtype>::FilterNet(const NetParameter& param,//将param中的不符合规则的层去掉,过滤后的层保存在param_filtered(比如prototxt文件中出现的include字段)
NetParameter* param_filtered) {
NetState net_state(param.state());
param_filtered->CopyFrom(param);//初始化param_filtered
param_filtered->clear_layer();//开始把它清空
for (int i = 0; i < param.layer_size(); ++i) {//逐层判断,include和exclude字段
const LayerParameter& layer_param = param.layer(i);
const string& layer_name = layer_param.name();
CHECK(layer_param.include_size() == 0 || layer_param.exclude_size() == 0)//include和exclude字段不能同时为空
<< "Specify either include rules or exclude rules; not both.";
bool layer_included = (layer_param.include_size() == 0);
for (int j = 0; layer_included && j < layer_param.exclude_size(); ++j) {//如果include字段为空(exclude字段就不为空),逐个判断是否满足exclude条件
if (StateMeetsRule(net_state, layer_param.exclude(j), layer_name)) {
layer_included = false;//满足exclude条件,则exclude,退出循环
}
}
for (int j = 0; !layer_included && j < layer_param.include_size(); ++j) {//如果exclude字段为空(include字段就不为空),逐个判断是否满足include条件
if (StateMeetsRule(net_state, layer_param.include(j), layer_name)) {
layer_included = true;//满足include条件,则include,退出循环
}
}
if (layer_included) {
param_filtered->add_layer()->CopyFrom(layer_param);//将include的层加入到param_filtered中,即为过滤后的层
}
}
}

template <typename Dtype>
bool Net<Dtype>::StateMeetsRule(const NetState& state,
const NetStateRule& rule, const string& layer_name) {//判断state符不符合rule条件
if (rule.has_phase()) {//如果阶段不一样,则不符合,比如train net中出现test
if (rule.phase() != state.phase()) {
LOG_IF(INFO, Caffe::root_solver())
<< "The NetState phase (" << state.phase()
<< ") differed from the phase (" << rule.phase()
<< ") specified by a rule in layer " << layer_name;
return false;
}
}
if (rule.has_min_level()) {//根据min_level
if (state.level() < rule.min_level()) {
LOG_IF(INFO, Caffe::root_solver())
<< "The NetState level (" << state.level()
<< ") is above the min_level (" << rule.min_level()
<< ") specified by a rule in layer " << layer_name;
return false;
}
}
if (rule.has_max_level()) {//根据maxin_level
if (state.level() > rule.max_level()) {
LOG_IF(INFO, Caffe::root_solver())
<< "The NetState level (" << state.level()
<< ") is above the max_level (" << rule.max_level()
<< ") specified by a rule in layer " << layer_name;
return false;
}
}
// Check whether the rule is broken due to stage. The NetState must
// contain ALL of the rule's stages to meet it.
for (int i = 0; i < rule.stage_size(); ++i) {
// Check that the NetState contains the rule's ith stage.
bool has_stage = false;
for (int j = 0; !has_stage && j < state.stage_size(); ++j) {
if (rule.stage(i) == state.stage(j)) { has_stage = true; }
}
if (!has_stage) {
LOG_IF(INFO, Caffe::root_solver())
<< "The NetState did not contain stage '" << rule.stage(i)
<< "' specified by a rule in layer " << layer_name;
return false;
}
}
// Check whether the rule is broken due to not_stage. The NetState must
// contain NONE of the rule's not_stages to meet it.
for (int i = 0; i < rule.not_stage_size(); ++i) {
// Check that the NetState contains the rule's ith not_stage.
bool has_stage = false;
for (int j = 0; !has_stage && j < state.stage_size(); ++j) {
if (rule.not_stage(i) == state.stage(j)) { has_stage = true; }
}
if (has_stage) {
LOG_IF(INFO, Caffe::root_solver())
<< "The NetState contained a not_stage '" << rule.not_stage(i)
<< "' specified by a rule in layer " << layer_name;
return false;
}
}
return true;
}

// Helper for Net::Init: add a new top blob to the net.
template <typename Dtype>
void Net<Dtype>::AppendTop(const NetParameter& param, const int layer_id,
const int top_id, set<string>* available_blobs,
map<string, int>* blob_name_to_idx) {//在某层的某个位置添加新的输出
shared_ptr<LayerParameter> layer_param(
new LayerParameter(param.layer(layer_id)));//为层具体分配内存,layer_param表示该层
const string& blob_name = (layer_param->top_size() > top_id) ?//获取名字,top装的是输出blob的名字
layer_param->top(top_id) : "(automatic)";
if (blob_name_to_idx && layer_param->bottom_size() > top_id &&
blob_name == layer_param->bottom(top_id)) {// In-place 计算,输入与输出是同一个blob
LOG_IF(INFO, Caffe::root_solver())
<< layer_param->name() << " -> " << blob_name << " (in-place)";
top_vecs_[layer_id].push_back(blobs_[(*blob_name_to_idx)[blob_name]].get());
top_id_vecs_[layer_id].push_back((*blob_name_to_idx)[blob_name]);
} else if (blob_name_to_idx &&
blob_name_to_idx->find(blob_name) != blob_name_to_idx->end()) {//如果没有进行In-place 计算,但有重复的blob则会报错
LOG(FATAL) << "Top blob '" << blob_name
<< "' produced by multiple sources.";
} else {//正常情况下
if (Caffe::root_solver()) {
LOG(INFO) << layer_param->name() << " -> " << blob_name;
}
shared_ptr<Blob<Dtype> > blob_pointer(new Blob<Dtype>());//开辟指针
const int blob_id = blobs_.size();//每个blob,都会压入一个blob指针,刚开始没有,blob_id为0,之后累加
blobs_.push_back(blob_pointer);//将指针压入blobs_
blob_names_.push_back(blob_name);//将名字压入blob_names_
blob_need_backward_.push_back(false);将该输出blob初始化为不需要backward,之后如果该层需要back,则会更新为true
if (blob_name_to_idx) { (*blob_name_to_idx)[blob_name] = blob_id; }//将blob_name与blob_id放到blob_name_to_idx里面
top_id_vecs_[layer_id].push_back(blob_id);//将blob_id放到该层的top_id_vecs_里面
top_vecs_[layer_id].push_back(blob_pointer.get());//将指向具体blob数据的指针放到该层的top_vecs_里面
}
if (available_blobs) { available_blobs->insert(blob_name); }//把该blob名字存到available_blobs
}
template <typename Dtype>
int Net<Dtype>::AppendBottom(const NetParameter& param, const int layer_id,//在某层的某个位置添加新的输入,类似于多输入单输出,你要再加个输入,实际上中间值的存储在appendtop的过程中,appendbottom过程只是调用,因为这一层的输入是上一层的输出
const int bottom_id, set<string>* available_blobs,
map<string, int>* blob_name_to_idx) {
const LayerParameter& layer_param = param.layer(layer_id);
const string& blob_name = layer_param.bottom(bottom_id);//抽取该输入blob的名字
if (available_blobs->find(blob_name) == available_blobs->end()) {//根据名字从所有可用blob中查找,找不到则报错,这一层的输入是上一层的输出
LOG(FATAL) << "Unknown bottom blob '" << blob_name << "' (layer '"
<< layer_param.name() << "', bottom index " << bottom_id << ")";
}
const int blob_id = (*blob_name_to_idx)[blob_name];//根据该blob的名字获取id
LOG_IF(INFO, Caffe::root_solver())
<< layer_names_[layer_id] << " <- " << blob_name;
bottom_vecs_[layer_id].push_back(blobs_[blob_id].get());//将指向实际数据的指针放进该层的bottom_vecs_里
bottom_id_vecs_[layer_id].push_back(blob_id);//将blob_id放进该层的bottom_id_vecs_里
available_blobs->erase(blob_name);//用过的blob则删除
bool need_backward = blob_need_backward_[blob_id];
// Check if the backpropagation on bottom_id should be skipped
if (layer_param.propagate_down_size() > 0) {
need_backward = layer_param.propagate_down(bottom_id);//根据propagate_down参数,判断该该输入是否需要backward
}
bottom_need_backward_[layer_id].push_back(need_backward);//将结果保存到bottom_need_backward_里面
return blob_id;
}

template <typename Dtype>
void Net<Dtype>::AppendParam(const NetParameter& param, const int layer_id,
const int param_id) {//加载层参数
const LayerParameter& layer_param = layers_[layer_id]->layer_param();
const int param_size = layer_param.param_size();
string param_name =
(param_size > param_id) ? layer_param.param(param_id).name() : "";//记录该种参数的name(在message ParamSpec中定义)
if (param_name.size()) {
param_display_names_.push_back(param_name);//如果不是空字符串,则压入param_display_names_
} else {
ostringstream param_display_name;//如果是空字符串,则将param_id转换为字符串压入param_display_names_
param_display_name << param_id;
param_display_names_.push_back(param_display_name.str());
}
const int net_param_id = params_.size();//net_param_id为参数全局id,开始为0.慢慢累加,param_id为每层的局部id
params_.push_back(layers_[layer_id]->blobs()[param_id]);//将该层第param_id参数指向的blobs()指针压入params_(params_.size()就会加1),其中blobs()装的是每个层的可学习参数
param_id_vecs_[layer_id].push_back(net_param_id);//将参数全局id压入param_id_vecs_
param_layer_indices_.push_back(make_pair(layer_id, param_id));//将当前层id与参数局部id一一对应,放进param_layer_indices_,下标为net_param_id
ParamSpec default_param_spec;
const ParamSpec* param_spec = (layer_param.param_size() > param_id) ?
&layer_param.param(param_id) : &default_param_spec;
if (!param_size || !param_name.size() || (param_name.size() &&
param_names_index_.find(param_name) == param_names_index_.end())) {//不共享参数的情况,这个时候 learnable_params_与params_存储一致
param_owners_.push_back(-1);//表示当前层就是该参数的"owner"
if (param_name.size()) {
param_names_index_[param_name] = net_param_id;//如果param_name不是空字符串,记录param_name及其owner id(net_param_id),存入param_names_index_里,为了在不同层之间共享参数,之后根据名字找到owner
}
const int learnable_param_id = learnable_params_.size();//learnable_params_用来装可学习参数
learnable_params_.push_back(params_[net_param_id].get());//指向可学习参数数据的指针(从整个网络来看,而不是每一层)
learnable_param_ids_.push_back(learnable_param_id);//可学习参数全局id
has_params_lr_.push_back(param_spec->has_lr_mult());//是否有学习率参数
has_params_decay_.push_back(param_spec->has_decay_mult());//是否有衰减率参数
params_lr_.push_back(param_spec->lr_mult());//获取学习率参数
params_weight_decay_.push_back(param_spec->decay_mult());//获取衰减率参数
} else {//共享参数的情况,根据param_name来共享,这个时候 learnable_params_与params_存储不一致,前者存指向不重复的参数的指针(重复不压入,只将id保存),后者保存所以层参数指针
const int owner_net_param_id = param_names_index_[param_name];//根据param_name找到其owner id
param_owners_.push_back(owner_net_param_id);//将该id作为该参数的owners
const pair<int, int>& owner_index =
param_layer_indices_[owner_net_param_id];//获取该owner 对应的层和局部参数id
const int owner_layer_id = owner_index.first;
const int owner_param_id = owner_index.second;
LOG_IF(INFO, Caffe::root_solver()) << "Sharing parameters '" << param_name
<< "' owned by "
<< "layer '" << layer_names_[owner_layer_id] << "', param "
<< "index " << owner_param_id;
Blob<Dtype>* this_blob = layers_[layer_id]->blobs()[param_id].get();//获取该层该参数对应的可学习参数指针
Blob<Dtype>* owner_blob =
layers_[owner_layer_id]->blobs()[owner_param_id].get();//获取该owner对应的可学习参数指针
const int param_size = layer_param.param_size();
if (param_size > param_id && (layer_param.param(param_id).share_mode() ==
ParamSpec_DimCheckMode_PERMISSIVE)) {//仅仅检查总数目是不是相等,相等才可以共享
CHECK_EQ(this_blob->count(), owner_blob->count())
<< "Cannot share param '" << param_name << "' owned by layer '"
<< layer_names_[owner_layer_id] << "' with layer '"
<< layer_names_[layer_id] << "'; count mismatch.  Owner layer param "
<< "shape is " << owner_blob->shape_string() << "; sharing layer "
<< "shape is " << this_blob->shape_string();
} else {
//严格检查每一维度是不是相等,相等才可以共享
CHECK(this_blob->shape() == owner_blob->shape())
<< "Cannot share param '" << param_name << "' owned by layer '"
<< layer_names_[owner_layer_id] << "' with layer '"
<< layer_names_[layer_id] << "'; shape mismatch.  Owner layer param "
<< "shape is " << owner_blob->shape_string() << "; sharing layer "
<< "expects shape " << this_blob->shape_string();
}
const int learnable_param_id = learnable_param_ids_[owner_net_param_id];
learnable_param_ids_.push_back(learnable_param_id);//将其owner id压入
if (param_spec->has_lr_mult()) {//有学习率参数
if (has_params_lr_[learnable_param_id]) {//其owner有学习率参数,要相等才可以
CHECK_EQ(param_spec->lr_mult(), params_lr_[learnable_param_id])
<< "Shared param '" << param_name << "' has mismatched lr_mult.";
} else {
has_params_lr_[learnable_param_id] = true;
params_lr_[learnable_param_id] = param_spec->lr_mult();//赋值
}
}
if (param_spec->has_decay_mult()) {//同上
if (has_params_decay_[learnable_param_id]) {
CHECK_EQ(param_spec->decay_mult(),
params_weight_decay_[learnable_param_id])
<< "Shared param '" << param_name << "' has mismatched decay_mult.";
} else {
has_params_decay_[learnable_param_id] = true;
params_weight_decay_[learnable_param_id] = param_spec->decay_mult();
}
}
}
}
template <typename Dtype>
Dtype Net<Dtype>::ForwardFromTo(int start, int end) {//从start 传播到 end 为止
CHECK_GE(start, 0);
CHECK_LT(end, layers_.size());
Dtype loss = 0;
for (int i = start; i <= end; ++i) {
// LOG(ERROR) << "Forwarding " << layer_names_[i];
Dtype layer_loss = layers_[i]->Forward(bottom_vecs_[i], top_vecs_[i]);//进入到层内部的Forward函数
loss += layer_loss;//将每一层的loss之和
if (debug_info_) { ForwardDebugInfo(i); }
}
return loss;
}
template <typename Dtype>
Dtype Net<Dtype>::ForwardFrom(int start) {//从start 传播到 最后
return ForwardFromTo(start, layers_.size() - 1);
}
template <typename Dtype>
Dtype Net<Dtype>::ForwardTo(int end) {//从开始传播到end
return ForwardFromTo(0, end);
}
template <typename Dtype>
const vector<Blob<Dtype>*>& Net<Dtype>::Forward(Dtype* loss) {//传播结束后获取网络输出并保存损失
if (loss != NULL) {
*loss = ForwardFromTo(0, layers_.size() - 1);
} else {
ForwardFromTo(0, layers_.size() - 1);
}
return net_output_blobs_;
}
template <typename Dtype>
const vector<Blob<Dtype>*>& Net<Dtype>::Forward(
const vector<Blob<Dtype>*> & bottom, Dtype* loss) {//不建议使用,用上面的
LOG_EVERY_N(WARNING, 1000) << "DEPRECATED: Forward(bottom, loss) "
<< "will be removed in a future version. Use Forward(loss).";
// Copy bottom to net bottoms
for (int i = 0; i < bottom.size(); ++i) {
net_input_blobs_[i]->CopyFrom(*bottom[i]);
}
return Forward(loss);
}
template <typename Dtype>
void Net<Dtype>::BackwardFromTo(int start, int end) {//进入到层内部的Backward,从start到end
CHECK_GE(end, 0);
CHECK_LT(start, layers_.size());
for (int i = start; i >= end; --i) {
if (layer_need_backward_[i]) {
layers_[i]->Backward(
top_vecs_[i], bottom_need_backward_[i], bottom_vecs_[i]);
if (debug_info_) { BackwardDebugInfo(i); }
}
}
}
template <typename Dtype>
void Net<Dtype>::ForwardDebugInfo(const int layer_id) {//打印给定层前向传播的调试信息
for (int top_id = 0; top_id < top_vecs_[layer_id].size(); ++top_id) {
const Blob<Dtype>& blob = *top_vecs_[layer_id][top_id];
const string& blob_name = blob_names_[top_id_vecs_[layer_id][top_id]];
const Dtype data_abs_val_mean = blob.asum_data() / blob.count();
LOG_IF(INFO, Caffe::root_solver())
<< "    [Forward] "
<< "Layer " << layer_names_[layer_id]
<< ", top blob " << blob_name
<< " data: " << data_abs_val_mean;
}
for (int param_id = 0; param_id < layers_[layer_id]->blobs().size();
++param_id) {
const Blob<Dtype>& blob = *layers_[layer_id]->blobs()[param_id];
const int net_param_id = param_id_vecs_[layer_id][param_id];
const string& blob_name = param_display_names_[net_param_id];
const Dtype data_abs_val_mean = blob.asum_data() / blob.count();
LOG_IF(INFO, Caffe::root_solver())
<< "    [Forward] "
<< "Layer " << layer_names_[layer_id]
<< ", param blob " << blob_name
<< " data: " << data_abs_val_mean;
}
}
template <typename Dtype>
void Net<Dtype>::BackwardDebugInfo(const int layer_id) {//打印给定层后向传播的调试信息
const vector<Blob<Dtype>*>& bottom_vec = bottom_vecs_[layer_id];
for (int bottom_id = 0; bottom_id < bottom_vec.size(); ++bottom_id) {
if (!bottom_need_backward_[layer_id][bottom_id]) { continue; }
const Blob<Dtype>& blob = *bottom_vec[bottom_id];
const string& blob_name = blob_names_[bottom_id_vecs_[layer_id][bottom_id]];
const Dtype diff_abs_val_mean = blob.asum_diff() / blob.count();
LOG_IF(INFO, Caffe::root_solver())
<< "    [Backward] "
<< "Layer " << layer_names_[layer_id]
<< ", bottom blob " << blob_name
<< " diff: " << diff_abs_val_mean;
}
for (int param_id = 0; param_id < layers_[layer_id]->blobs().size();
++param_id) {
if (!layers_[layer_id]->param_propagate_down(param_id)) { continue; }
const Blob<Dtype>& blob = *layers_[layer_id]->blobs()[param_id];
const Dtype diff_abs_val_mean = blob.asum_diff() / blob.count();
LOG_IF(INFO, Caffe::root_solver())
<< "    [Backward] "
<< "Layer " << layer_names_[layer_id]
<< ", param blob " << param_id
<< " diff: " << diff_abs_val_mean;
}
}
template <typename Dtype>
void Net<Dtype>::UpdateDebugInfo(const int param_id) {//打印更新过程的调试信息
const Blob<Dtype>& blob = *params_[param_id];
const int param_owner = param_owners_[param_id];
const string& layer_name = layer_names_[param_layer_indices_[param_id].first];
const string& param_display_name = param_display_names_[param_id];
const Dtype diff_abs_val_mean = blob.asum_diff() / blob.count();
if (param_owner < 0) {
const Dtype data_abs_val_mean = blob.asum_data() / blob.count();
LOG_IF(INFO, Caffe::root_solver())
<< "    [Update] Layer " << layer_name
<< ", param " << param_display_name
<< " data: " << data_abs_val_mean
<< "; diff: " << diff_abs_val_mean;
} else {
const string& owner_layer_name =
layer_names_[param_layer_indices_[param_owner].first];
LOG_IF(INFO, Caffe::root_solver())
<< "    [Update] Layer " << layer_name
<< ", param blob " << param_display_name
<< " (owned by layer " << owner_layer_name << ", " << "param "
<< param_display_names_[param_owners_[param_id]] << ")"
<< " diff: " << diff_abs_val_mean;
}
}
template <typename Dtype>
void Net<Dtype>::ShareTrainedLayersWith(const Net* other) {//从其他网络中加载参数
int num_source_layers = other->layers().size();
for (int i = 0; i < num_source_layers; ++i) {
Layer<Dtype>* source_layer = other->layers()[i].get();
const string& source_layer_name = other->layer_names()[i];
int target_layer_id = 0;
while (target_layer_id != layer_names_.size() &&
layer_names_[target_layer_id] != source_layer_name) {
++target_layer_id;
}
if (target_layer_id == layer_names_.size()) {
LOG(INFO) << "Ignoring source layer " << source_layer_name;
continue;
}
DLOG(INFO) << "Copying source layer " << source_layer_name;
vector<shared_ptr<Blob<Dtype> > >& target_blobs =
layers_[target_layer_id]->blobs();
CHECK_EQ(target_blobs.size(), source_layer->blobs().size())
<< "Incompatible number of blobs for layer " << source_layer_name;
for (int j = 0; j < target_blobs.size(); ++j) {
Blob<Dtype>* source_blob = source_layer->blobs()[j].get();
CHECK(target_blobs[j]->shape() == source_blob->shape())
<< "Cannot share param " << j << " weights from layer '"
<< source_layer_name << "'; shape mismatch.  Source param shape is "
<< source_blob->shape_string() << "; target param shape is "
<< target_blobs[j]->shape_string();
target_blobs[j]->ShareData(*source_blob);
}
}
}
template <typename Dtype>
void Net<Dtype>::BackwardFrom(int start) {
BackwardFromTo(start, 0);
}
template <typename Dtype>
void Net<Dtype>::BackwardTo(int end) {
BackwardFromTo(layers_.size() - 1, end);
}
template <typename Dtype>
void Net<Dtype>::Backward() {
BackwardFromTo(layers_.size() - 1, 0);
if (debug_info_) {
Dtype asum_data = 0, asum_diff = 0, sumsq_data = 0, sumsq_diff = 0;
for (int i = 0; i < learnable_params_.size(); ++i) {
asum_data += learnable_params_[i]->asum_data();
asum_diff += learnable_params_[i]->asum_diff();
sumsq_data += learnable_params_[i]->sumsq_data();
sumsq_diff += learnable_params_[i]->sumsq_diff();
}
const Dtype l2norm_data = std::sqrt(sumsq_data);
const Dtype l2norm_diff = std::sqrt(sumsq_diff);
LOG(ERROR) << "    [Backward] All net params (data, diff): "
<< "L1 norm = (" << asum_data << ", " << asum_diff << "); "
<< "L2 norm = (" << l2norm_data << ", " << l2norm_diff << ")";
}
}
template <typename Dtype>
void Net<Dtype>::Reshape() {//塑形
for (int i = 0; i < layers_.size(); ++i) {
layers_[i]->Reshape(bottom_vecs_[i], top_vecs_[i]);
}
}
template <typename Dtype>
void Net<Dtype>::CopyTrainedLayersFrom(const NetParameter& param) {//从给定网络参数中加载参数
int num_source_layers = param.layer_size();//给定网络中层的数量
for (int i = 0; i < num_source_layers; ++i) {//对于每一层
const LayerParameter& source_layer = param.layer(i);
const string& source_layer_name = source_layer.name();//层的名字
int target_layer_id = 0;
while (target_layer_id != layer_names_.size() &&
layer_names_[target_layer_id] != source_layer_name) {//如果网络中没有给定网络中的层,则进行下一层寻找
++target_layer_id;
}
if (target_layer_id == layer_names_.size()) {
LOG(INFO) << "Ignoring source layer " << source_layer_name;
continue;
}
DLOG(INFO) << "Copying source layer " << source_layer_name;
vector<shared_ptr<Blob<Dtype> > >& target_blobs =
layers_[target_layer_id]->blobs();
CHECK_EQ(target_blobs.size(), source_layer.blobs_size())
<< "Incompatible number of blobs for layer " << source_layer_name;
for (int j = 0; j < target_blobs.size(); ++j) {
if (!target_blobs[j]->ShapeEquals(source_layer.blobs(j))) {
Blob<Dtype> source_blob;
const bool kReshape = true;
source_blob.FromProto(source_layer.blobs(j), kReshape);
LOG(FATAL) << "Cannot copy param " << j << " weights from layer '"
<< source_layer_name << "'; shape mismatch.  Source param shape is "
<< source_blob.shape_string() << "; target param shape is "
<< target_blobs[j]->shape_string() << ". "
<< "To learn this layer's parameters from scratch rather than "
<< "copying from a saved net, rename the layer.";
}
const bool kReshape = false;
target_blobs[j]->FromProto(source_layer.blobs(j), kReshape);
}
}
}
template <typename Dtype>
void Net<Dtype>::CopyTrainedLayersFrom(const string trained_filename) {//从文件中加载参数,根据文件类型,调用各自函数加载参数
if (trained_filename.size() >= 3 &&
trained_filename.compare(trained_filename.size() - 3, 3, ".h5") == 0) {
CopyTrainedLayersFromHDF5(trained_filename);
} else {
CopyTrainedLayersFromBinaryProto(trained_filename);
}
}
template <typename Dtype>
void Net<Dtype>::CopyTrainedLayersFromBinaryProto(//从proto文件中加载参数
const string trained_filename) {
NetParameter param;
ReadNetParamsFromBinaryFileOrDie(trained_filename, ¶m);
CopyTrainedLayersFrom(param);
}
template <typename Dtype>
void Net<Dtype>::CopyTrainedLayersFromHDF5(const string trained_filename) {//从HDF5文件中加载参数
hid_t file_hid = H5Fopen(trained_filename.c_str(), H5F_ACC_RDONLY,
H5P_DEFAULT);
CHECK_GE(file_hid, 0) << "Couldn't open " << trained_filename;
hid_t data_hid = H5Gopen2(file_hid, "data", H5P_DEFAULT);
CHECK_GE(data_hid, 0) << "Error reading weights from " << trained_filename;
int num_layers = hdf5_get_num_links(data_hid);
for (int i = 0; i < num_layers; ++i) {
string source_layer_name = hdf5_get_name_by_idx(data_hid, i);
if (!layer_names_index_.count(source_layer_name)) {
LOG(INFO) << "Ignoring source layer " << source_layer_name;
continue;
}
int target_layer_id = layer_names_index_[source_layer_name];
DLOG(INFO) << "Copying source layer " << source_layer_name;
vector<shared_ptr<Blob<Dtype> > >& target_blobs =
layers_[target_layer_id]->blobs();
hid_t layer_hid = H5Gopen2(data_hid, source_layer_name.c_str(),
H5P_DEFAULT);
CHECK_GE(layer_hid, 0)
<< "Error reading weights from " << trained_filename;
// Check that source layer doesn't have more params than target layer
int num_source_params = hdf5_get_num_links(layer_hid);
CHECK_LE(num_source_params, target_blobs.size())
<< "Incompatible number of blobs for layer " << source_layer_name;
for (int j = 0; j < target_blobs.size(); ++j) {
ostringstream oss;
oss << j;
string dataset_name = oss.str();
int target_net_param_id = param_id_vecs_[target_layer_id][j];
if (!H5Lexists(layer_hid, dataset_name.c_str(), H5P_DEFAULT)) {
// Target param doesn't exist in source weights...
if (param_owners_[target_net_param_id] != -1) {
// ...but it's weight-shared in target, so that's fine.
continue;
} else {
LOG(FATAL) << "Incompatible number of blobs for layer "
<< source_layer_name;
}
}
hdf5_load_nd_dataset(layer_hid, dataset_name.c_str(), 0, kMaxBlobAxes,
target_blobs[j].get());
}
H5Gclose(layer_hid);
}
H5Gclose(data_hid);
H5Fclose(file_hid);
}
template <typename Dtype>
void Net<Dtype>::ToProto(NetParameter* param, bool write_diff) const {//序列化到proto buf
param->Clear();
param->set_name(name_);
DLOG(INFO) << "Serializing " << layers_.size() << " layers";
for (int i = 0; i < layers_.size(); ++i) {
LayerParameter* layer_param = param->add_layer();
layers_[i]->ToProto(layer_param, write_diff);
}
}
template <typename Dtype>
void Net<Dtype>::ToHDF5(const string& filename, bool write_diff) const {//转换为HDF5文件
hid_t file_hid = H5Fcreate(filename.c_str(), H5F_ACC_TRUNC, H5P_DEFAULT,
H5P_DEFAULT);
CHECK_GE(file_hid, 0)
<< "Couldn't open " << filename << " to save weights.";
hid_t data_hid = H5Gcreate2(file_hid, "data", H5P_DEFAULT, H5P_DEFAULT,
H5P_DEFAULT);
CHECK_GE(data_hid, 0) << "Error saving weights to " << filename << ".";
hid_t diff_hid = -1;
if (write_diff) {
diff_hid = H5Gcreate2(file_hid, "diff", H5P_DEFAULT, H5P_DEFAULT,
H5P_DEFAULT);
CHECK_GE(diff_hid, 0) << "Error saving weights to " << filename << ".";
}
for (int layer_id = 0; layer_id < layers_.size(); ++layer_id) {
const LayerParameter& layer_param = layers_[layer_id]->layer_param();
string layer_name = layer_param.name();
hid_t layer_data_hid = H5Gcreate2(data_hid, layer_name.c_str(),
H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
CHECK_GE(layer_data_hid, 0)
<< "Error saving weights to " << filename << ".";
hid_t layer_diff_hid = -1;
if (write_diff) {
layer_diff_hid = H5Gcreate2(diff_hid, layer_name.c_str(),
H5P_DEFAULT, H5P_DEFAULT, H5P_DEFAULT);
CHECK_GE(layer_diff_hid, 0)
<< "Error saving weights to " << filename << ".";
}
int num_params = layers_[layer_id]->blobs().size();
for (int param_id = 0; param_id < num_params; ++param_id) {
ostringstream dataset_name;
dataset_name << param_id;
const int net_param_id = param_id_vecs_[layer_id][param_id];
if (param_owners_[net_param_id] == -1) {
// Only save params that own themselves
hdf5_save_nd_dataset<Dtype>(layer_data_hid, dataset_name.str(),
*params_[net_param_id]);
}
if (write_diff) {
// Write diffs regardless of weight-sharing
hdf5_save_nd_dataset<Dtype>(layer_diff_hid, dataset_name.str(),
*params_[net_param_id], true);
}
}
H5Gclose(layer_data_hid);
if (write_diff) {
H5Gclose(layer_diff_hid);
}
}
H5Gclose(data_hid);
if (write_diff) {
H5Gclose(diff_hid);
}
H5Fclose(file_hid);
}
template <typename Dtype>
void Net<Dtype>::Update() {//更新可学习参数
for (int i = 0; i < learnable_params_.size(); ++i) {
learnable_params_[i]->Update();
}
}
template <typename Dtype>
void Net<Dtype>::ClearParamDiffs() {//重置参数梯度
for (int i = 0; i < learnable_params_.size(); ++i) {
Blob<Dtype>* blob = learnable_params_[i];
switch (Caffe::mode()) {
case Caffe::CPU:
caffe_set(blob->count(), static_cast<Dtype>(0),
blob->mutable_cpu_diff());
break;
case Caffe::GPU:
#ifndef CPU_ONLY
caffe_gpu_set(blob->count(), static_cast<Dtype>(0),
blob->mutable_gpu_diff());
#else
NO_GPU;
#endif
break;
}
}
}
template <typename Dtype>
void Net<Dtype>::ShareWeights() {//共享权重
for (int i = 0; i < params_.size(); ++i) {
if (param_owners_[i] < 0) { continue; }//该参数属于当前层,不需要共享
params_[i]->ShareData(*params_[param_owners_[i]]);//共享参数数据
params_[i]->ShareDiff(*params_[param_owners_[i]]);//共享参数梯度
}
}
template <typename Dtype>
bool Net<Dtype>::has_blob(const string& blob_name) const {//根据名字查看是否包括该blob
return blob_names_index_.find(blob_name) != blob_names_index_.end();
}
template <typename Dtype>
const shared_ptr<Blob<Dtype> > Net<Dtype>::blob_by_name(
const string& blob_name) const {//如果有该blob的话,则获取该blob的指针
shared_ptr<Blob<Dtype> > blob_ptr;
if (has_blob(blob_name)) {
blob_ptr = blobs_[blob_names_index_.find(blob_name)->second];
} else {
blob_ptr.reset((Blob<Dtype>*)(NULL));
LOG(WARNING) << "Unknown blob name " << blob_name;
}
return blob_ptr;
}
template <typename Dtype>
bool Net<Dtype>::has_layer(const string& layer_name) const {//根据层名查看网络中是否有该层
return layer_names_index_.find(layer_name) != layer_names_index_.end();
}
template <typename Dtype>
const shared_ptr<Layer<Dtype> > Net<Dtype>::layer_by_name(
const string& layer_name) const {//如果有的话,则返回指向该层的指针
shared_ptr<Layer<Dtype> > layer_ptr;
if (has_layer(layer_name)) {
layer_ptr = layers_[layer_names_index_.find(layer_name)->second];
} else {
layer_ptr.reset((Layer<Dtype>*)(NULL));
LOG(WARNING) << "Unknown layer name " << layer_name;
}
return layer_ptr;
}

INSTANTIATE_CLASS(Net);

}  // namespace caffe