shared_ptr<Blob<float> > device::Buffer(int id) {
if (buff_f_.size() <= id) {
shared_ptr<Blob<float> > blob_pointer(new Blob<float>(this));
buff_f_.push_back(blob_pointer);
}
return buff_f_[id];
}
shared_ptr<Blob<double> > DeviceContext::Buffer(int id) {
if (buff_d_.size() <= id) {
shared_ptr<Blob<double> > blob_pointer(new Blob<double>(this));
buff_d_.push_back(blob_pointer);
}
return buff_d_[id];
}
char *buffer_read_string(uint32_t len)
{
char *s;
blob_free(blob_pool.size);
s = blob_pointer(blob_alloc(len + 1));
s[fread(s, 1, len, infile)] = '\0';
return ferror(infile) ? NULL : s;
}
void AffinityLayer<Dtype>::Reshape(const vector<Blob<Dtype>*>& bottom,
const vector<Blob<Dtype>*>& top) {
min_index_.clear();
for (int bidx = 0; bidx < bottom.size(); ++bidx) {
// 1, #edges, height, width
top[bidx]->Reshape(1, 2, bottom[bidx]->height(), bottom[bidx]->width());
shared_ptr<Blob<Dtype> > blob_pointer(
new Blob<Dtype>(this->device_context()));
min_index_.push_back(blob_pointer);
// 1, #edges, height, width
min_index_[bidx]->Reshape(1, 2, bottom[bidx]->height(),
bottom[bidx]->width());
}
}
Dtype ApolloNet<Dtype>::ForwardLayer(const string& layer_param_string, const string& runtime_param_string) {
/* This function will
* 1) Check if the layer name is in the cache
* 2) Create the layer if it is new
* 3) Set up the top blobs
* 4) Set up the bottom blobs
* 5) Set up the parameters
* 6) Call the Forward function */
LayerParameter active_layer_param;
RuntimeParameter runtime_param;
ASSERT(runtime_param.ParseFromString(runtime_param_string), "");
ASSERT(active_layer_param.ParseFromString(layer_param_string), "");
ASSERT(active_layer_param.has_name(), "");
const string& layer_name = active_layer_param.name();
shared_ptr<Layer<Dtype> > layer;
const bool new_layer = layers_map_.find(layer_name) == layers_map_.end();
if (new_layer) {
layer = LayerRegistry<Dtype>::CreateLayer(active_layer_param);;
LOG(INFO) << "Creating Layer " << layer_name;
LOG(INFO) << active_layer_param.DebugString();
layers_map_[layer_name] = layer;
active_layers_set_.insert(layer_name);
} else {
layer = layers_map_[layer_name];
std::pair<set<string>::iterator,bool> ret = active_layers_set_.insert(layer_name);
ASSERT(ret.second, "Layer with name '" << layer_name << "' is already used");
ASSERT(layer->layer_param().type() == active_layer_param.type(),
"WARNING: layer with name '" << active_layer_param.name() << "' and different type already exists.");
}
layer->set_runtime_param(runtime_param);
active_layers_vec_.push_back(layer_name);
vector<Blob<Dtype>*> bottom_vec;
vector<Blob<Dtype>*> top_vec;
const vector<string>& bottom_names = bottom_blob_names_[layer_name];
bool reset_bottoms = active_layer_param.bottom_size() != bottom_names.size();
for (int bottom_id = 0; bottom_id < active_layer_param.bottom_size(); ++bottom_id) {
const string& blob_name = active_layer_param.bottom(bottom_id);
ASSERT(tops_.find(blob_name) != tops_.end(),
"Could not find bottom: '" << blob_name << "' for layer: " << layer_name);
if (bottom_names.size() > bottom_id && bottom_names[bottom_id] != blob_name) { reset_bottoms = true; }
}
if (new_layer || reset_bottoms) {
// layer is new, or it's list of bottoms has changed. Reset the bottom blobs
bottom_blobs_[layer_name].clear();
bottom_blob_names_[layer_name].clear();
for (int bottom_id = 0; bottom_id < active_layer_param.bottom_size(); ++bottom_id) {
const string& blob_name = active_layer_param.bottom(bottom_id);
bottom_blob_names_[layer_name].push_back(blob_name);
shared_ptr<Blob<Dtype> > top_blob = tops_[blob_name];
shared_ptr<Blob<Dtype> > bottom_blob(new Blob<Dtype>(top_blob->shape()));
bottom_blob->ShareData(*top_blob);
if (!layer->overwrites_bottom_diffs()) {
// if layer accumulates delta rather than overwriting, we can save memory
bottom_blob->ShareDiff(*top_blob);
}
bottom_blobs_[layer_name].push_back(bottom_blob);
}
layer->reset_bottoms(bottom_blob_names_[layer_name]);
} else {
// Reshape bottom_blobs to match their respective top blobs
for (int bottom_id = 0; bottom_id < active_layer_param.bottom_size(); ++bottom_id) {
const string& blob_name = active_layer_param.bottom(bottom_id);
shared_ptr<Blob<Dtype> > top_blob = tops_[blob_name];
shared_ptr<Blob<Dtype> > bottom_blob = bottom_blobs_[layer_name][bottom_id];
bottom_blob->ReshapeLike(*top_blob);
}
}
for (int bottom_id = 0; bottom_id < active_layer_param.bottom_size(); ++bottom_id) {
bottom_vec.push_back(bottom_blobs_[layer_name][bottom_id].get());
}
for (int top_id = 0; top_id < active_layer_param.top_size(); ++top_id) {
const string& blob_name = active_layer_param.top(top_id);
if (tops_.find(blob_name) == tops_.end()) {
shared_ptr<Blob<Dtype> > blob_pointer(new Blob<Dtype>());
tops_[blob_name] = blob_pointer;
}
Blob<Dtype>* top_blob = tops_[blob_name].get();
top_vec.push_back(top_blob);
if (top_blob->DiffInitialized() && !layer->is_loss()) {
// Zero out top_diffs, except for loss blobs, which never change
top_blob->SetDiffValues(0.);
}
}
if (new_layer) {
layer->SetUp(bottom_vec, top_vec);
AddLayerParams(layer);
}
for (int param_id = 0; param_id < layer->param_names().size(); ++param_id) {
const string& param_name = layer->param_names()[param_id];
active_params_set_.insert(param_name);
}
Dtype loss = 0;
layer->set_phase(phase_);
loss = layer->Forward(bottom_vec, top_vec);
return loss;
}
void Net<Dtype>::Init(const NetParameter& in_param) {
// Create a copy of in_param with splits added where necessary.
NetParameter param;
InsertSplits(in_param, ¶m);
// Basically, build all the layers and set up its connections.
name_ = param.name();
map<string, int> blob_name_to_idx;
set<string> available_blobs;
int num_layers = param.layers_size();
CHECK_EQ(param.input_size() * 4, param.input_dim_size())
<< "Incorrect bottom blob dimension specifications.";
size_t memory_used = 0;
// set the input blobs
for (int i = 0; i < param.input_size(); ++i) {
const string& blob_name = param.input(i);
shared_ptr<Blob<Dtype> > blob_pointer(
new Blob<Dtype>(param.input_dim(i * 4),
param.input_dim(i * 4 + 1),
param.input_dim(i * 4 + 2),
param.input_dim(i * 4 + 3)));
blobs_.push_back(blob_pointer);
blob_names_.push_back(blob_name);
blob_need_backward_.push_back(param.force_backward());
net_input_blob_indices_.push_back(i);
net_input_blobs_.push_back(blob_pointer.get());
blob_name_to_idx[blob_name] = i;
available_blobs.insert(blob_name);
memory_used += blob_pointer->count();
}
DLOG(INFO) << "Memory required for Data" << memory_used*sizeof(Dtype);
// For each layer, set up their input and output
bottom_vecs_.resize(param.layers_size());
top_vecs_.resize(param.layers_size());
bottom_id_vecs_.resize(param.layers_size());
top_id_vecs_.resize(param.layers_size());
for (int i = 0; i < param.layers_size(); ++i) {
bool in_place = false;
const LayerParameter& layer_param = param.layers(i);
layers_.push_back(shared_ptr<Layer<Dtype> >(GetLayer<Dtype>(layer_param)));
layer_names_.push_back(layer_param.name());
LOG(INFO) << "Creating Layer " << layer_param.name();
bool need_backward = param.force_backward();
// Figure out this layer's input
for (int j = 0; j < layer_param.bottom_size(); ++j) {
const string& blob_name = layer_param.bottom(j);
const int blob_id = blob_name_to_idx[blob_name];
if (available_blobs.find(blob_name) == available_blobs.end()) {
LOG(FATAL) << "Unknown blob input " << blob_name <<
" to layer" << j;
}
LOG(INFO) << layer_param.name() << " <- " << blob_name;
bottom_vecs_[i].push_back(blobs_[blob_id].get());
bottom_id_vecs_[i].push_back(blob_id);
// If a blob needs backward, this layer should provide it.
need_backward |= blob_need_backward_[blob_id];
available_blobs.erase(blob_name);
}
// Figure out this layer's output
for (int j = 0; j < layer_param.top_size(); ++j) {
const string& blob_name = layer_param.top(j);
// Check if we are doing in-place computation
if (layer_param.bottom_size() > j &&
blob_name == layer_param.bottom(j)) {
// In-place computation
LOG(INFO) << layer_param.name() << " -> " << blob_name << " (in-place)";
in_place = true;
available_blobs.insert(blob_name);
top_vecs_[i].push_back(
blobs_[blob_name_to_idx[blob_name]].get());
top_id_vecs_[i].push_back(blob_name_to_idx[blob_name]);
} else if (blob_name_to_idx.find(blob_name) != blob_name_to_idx.end()) {
// If we are not doing in-place computation but has duplicated blobs,
// raise an error.
LOG(FATAL) << "Duplicate blobs produced by multiple sources.";
} else {
// Normal output.
LOG(INFO) << layer_param.name() << " -> " << blob_name;
shared_ptr<Blob<Dtype> > blob_pointer(new Blob<Dtype>());
blobs_.push_back(blob_pointer);
blob_names_.push_back(blob_name);
blob_need_backward_.push_back(param.force_backward());
blob_name_to_idx[blob_name] = blob_names_.size() - 1;
available_blobs.insert(blob_name);
top_vecs_[i].push_back(blobs_[blob_names_.size() - 1].get());
top_id_vecs_[i].push_back(blob_names_.size() - 1);
}
}
// After this layer is connected, set it up.
//LOG(INFO) << "Setting up " << layer_names_[i];
layers_[i]->SetUp(bottom_vecs_[i], &(top_vecs_[i]));
for (int topid = 0; topid < top_vecs_[i].size(); ++topid) {
LOG(INFO) << "Top shape: " << top_vecs_[i][topid]->num() << " "
<< top_vecs_[i][topid]->channels() << " "
<< top_vecs_[i][topid]->height() << " "
<< top_vecs_[i][topid]->width() << " ("
<< top_vecs_[i][topid]->count() << ")";
if (!in_place)
memory_used += top_vecs_[i][topid]->count();
}
DLOG(INFO) << "Memory required for Data " << memory_used*sizeof(Dtype);
// blobs: 0# weights, 1# bias term; blob_lr: 1# learning rate for weights, 2# learning rate for bias
int blobs_lr_size = layers_[i]->layer_param().blobs_lr_size();
CHECK(blobs_lr_size == layers_[i]->blobs().size() || blobs_lr_size == 0) // 0, 1, 2
<< "Incorrect blobs lr size: should be either 0 or the same as "
"the number of the layer's parameter blobs.";
if (blobs_lr_size) {
// Check if this layer needs backward operation itself
for (int j = 0; j < blobs_lr_size; ++j) {
need_backward |= (layers_[i]->layer_param().blobs_lr(j) > 0);
}
} else if (layers_[i]->blobs().size()) {
// catch: if a layer param does not specify blobs_lr, we should assume the
// learning rate to be 1. Thus we will need to perform backward.
need_backward = true;
}
// Finally, set the backward flag
layer_need_backward_.push_back(need_backward);
if (need_backward) {
LOG(INFO) << layer_names_[i] << " needs backward computation.";
for (int j = 0; j < top_id_vecs_[i].size(); ++j) {
blob_need_backward_[top_id_vecs_[i][j]] = true;
}
} else {
LOG(INFO) << layer_names_[i] << " does not need backward computation.";
}
}
// In the end, all remaining blobs are considered output blobs.
for (set<string>::iterator it = available_blobs.begin();
it != available_blobs.end(); ++it) {
LOG(INFO) << "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 i = 0; i < blob_names_.size(); ++i) {
blob_names_index_[blob_names_[i]] = i;
}
for (size_t i = 0; i < layer_names_.size(); ++i) {
layer_names_index_[layer_names_[i]] = i;
}
GetLearningRateAndWeightDecay();
LOG(INFO) << "Network initialization done.";
LOG(INFO) << "Memory required for Data " << memory_used*sizeof(Dtype);
}
void Net<Dtype>::Init(const NetParameter& param) {
// Basically, build all the layers and set up its connections.
name_ = param.name();
map<string, int> blob_name_to_idx;
set<string> available_blobs;
int num_layers = param.layers_size();
CHECK_EQ(param.input_size() * 4, param.input_dim_size())
<< "Incorrect bottom blob dimension specifications.";
// set the input blobs
for (int i = 0; i < param.input_size(); ++i) {
const string& blob_name = param.input(i);
shared_ptr<Blob<Dtype> > blob_pointer(
new Blob<Dtype>(param.input_dim(i * 4),
param.input_dim(i * 4 + 1),
param.input_dim(i * 4 + 2),
param.input_dim(i * 4 + 3)));
blobs_.push_back(blob_pointer);
blob_names_.push_back(blob_name);
blob_need_backward_.push_back(param.force_backward());
net_input_blob_indices_.push_back(i);
net_input_blobs_.push_back(blob_pointer.get());
blob_name_to_idx[blob_name] = i;
available_blobs.insert(blob_name);
}
// For each layer, set up their input and output
bottom_vecs_.resize(param.layers_size());
top_vecs_.resize(param.layers_size());
bottom_id_vecs_.resize(param.layers_size());
top_id_vecs_.resize(param.layers_size());
for (int i = 0; i < param.layers_size(); ++i) {
const LayerConnection& layer_connection = param.layers(i);
const LayerParameter& layer_param = layer_connection.layer();
layers_.push_back(shared_ptr<Layer<Dtype> >(GetLayer<Dtype>(layer_param)));
layer_names_.push_back(layer_param.name());
LOG(INFO) << "Creating Layer " << layer_param.name();
bool need_backward = param.force_backward();
// Figure out this layer's input and output
for (int j = 0; j < layer_connection.bottom_size(); ++j) {
const string& blob_name = layer_connection.bottom(j);
const int blob_id = blob_name_to_idx[blob_name];
if (available_blobs.find(blob_name) == available_blobs.end()) {
LOG(FATAL) << "Unknown blob input " << blob_name <<
" to layer" << j;
}
LOG(INFO) << layer_param.name() << " <- " << blob_name;
bottom_vecs_[i].push_back(
blobs_[blob_id].get());
bottom_id_vecs_[i].push_back(blob_id);
// If a blob needs backward, this layer should provide it.
need_backward |= blob_need_backward_[blob_id];
available_blobs.erase(blob_name);
}
for (int j = 0; j < layer_connection.top_size(); ++j) {
const string& blob_name = layer_connection.top(j);
// Check if we are doing in-place computation
if (layer_connection.bottom_size() > j &&
blob_name == layer_connection.bottom(j)) {
// In-place computation
LOG(INFO) << layer_param.name() << " -> " << blob_name << " (in-place)";
available_blobs.insert(blob_name);
top_vecs_[i].push_back(
blobs_[blob_name_to_idx[blob_name]].get());
top_id_vecs_[i].push_back(blob_name_to_idx[blob_name]);
} else if (blob_name_to_idx.find(blob_name) != blob_name_to_idx.end()) {
// If we are not doing in-place computation but has duplicated blobs,
// raise an error.
LOG(FATAL) << "Duplicate blobs produced by multiple sources.";
} else {
// Normal output.
LOG(INFO) << layer_param.name() << " -> " << blob_name;
shared_ptr<Blob<Dtype> > blob_pointer(new Blob<Dtype>());
blobs_.push_back(blob_pointer);
blob_names_.push_back(blob_name);
blob_need_backward_.push_back(param.force_backward());
blob_name_to_idx[blob_name] = blob_names_.size() - 1;
available_blobs.insert(blob_name);
top_vecs_[i].push_back(blobs_[blob_names_.size() - 1].get());
top_id_vecs_[i].push_back(blob_names_.size() - 1);
}
}
// After this layer is connected, set it up.
// LOG(INFO) << "Setting up " << layer_names_[i];
layers_[i]->SetUp(bottom_vecs_[i], &top_vecs_[i]);
for (int topid = 0; topid < top_vecs_[i].size(); ++topid) {
LOG(INFO) << "Top shape: " << top_vecs_[i][topid]->channels() << " "
<< top_vecs_[i][topid]->height() << " "
<< top_vecs_[i][topid]->width();
}
// Check if this layer needs backward operation itself
for (int j = 0; j < layers_[i]->layer_param().blobs_lr_size(); ++j) {
need_backward |= (layers_[i]->layer_param().blobs_lr(j) > 0);
}
// Finally, set the backward flag
layer_need_backward_.push_back(need_backward);
if (need_backward) {
LOG(INFO) << layer_names_[i] << " needs backward computation.";
for (int j = 0; j < top_id_vecs_[i].size(); ++j) {
blob_need_backward_[top_id_vecs_[i][j]] = true;
}
} else {
LOG(INFO) << layer_names_[i] << " does not need backward computation.";
}
}
// In the end, all remaining blobs are considered output blobs.
for (set<string>::iterator it = available_blobs.begin();
it != available_blobs.end(); ++it) {
LOG(INFO) << "This network produces output " << *it;
net_output_blobs_.push_back(blobs_[blob_name_to_idx[*it]].get());
}
GetLearningRateAndWeightDecay();
LOG(INFO) << "Network initialization done.";
}
void TIConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype> *> &bottom,
vector<Blob<Dtype> *> *top) {
CHECK_EQ(bottom.size(), 1) << "TIConv Layer takes a single blob as input.";
CHECK_EQ(top->size(), 1) << "TIConv Layer takes a single blob as output.";
// NUM_T_ includes includes Identity transform
this->NUM_T_ = this->layer_param_.transformations_size();
LOG(INFO) << "TIConvolution layer using " << NUM_T_ << " transformations "
<< this->layer_param_.name() << " using interpolation: "
<< this->layer_param_.transformations(0).interp();
TransParameter tparam;
for (int t = 0; t < this->NUM_T_; ++t) {
tparam = this->layer_param_.transformations(t);
LOG(INFO) << " T" << t << " : "
<< " sc: " << tparam.scale() << ", rot: " << tparam.rotation();
}
if (Caffe::phase() == Caffe::TRAIN)
LOG(INFO) << " Creating Upsampling Layer in " << this->layer_param_.name();
this->up_layer_ = new UpsamplingLayer<Dtype>(this->layer_param_);
// The bottom that net provides is the bottom of UP layer
// for T transformations, need to make blob for each
for (int t = 0; t < this->NUM_T_; ++t) {
shared_ptr<Blob<Dtype> > blob_pointer(new Blob<Dtype>());
activations_.push_back(blob_pointer);
// and add them to up_top_vec_
up_top_vec_.push_back(blob_pointer.get());
}
// Setup UP layer, print shape
this->up_layer_->SetUp(bottom, &up_top_vec_);
if (Caffe::phase() == Caffe::TRAIN) {
for (int i = 0; i < up_top_vec_.size(); ++i) {
LOG(INFO) << " Top shape: " << up_top_vec_[i]->channels() << " "
<< up_top_vec_[i]->height() << " " << up_top_vec_[i]->width();
}
}
// Tied Conv
if (Caffe::phase() == Caffe::TRAIN)
LOG(INFO) << " Creating TiedConv Layer in " << this->layer_param_.name();
this->tiedconv_layer_ = new TiedConvolutionLayer<Dtype>(this->layer_param_);
// make new top blobs and add them to tiedconv_top_vec_
for (int t = 0; t < this->NUM_T_; ++t) {
shared_ptr<Blob<Dtype> > blob_pointer(new Blob<Dtype>());
activations_.push_back(blob_pointer);
tiedconv_top_vec_.push_back(blob_pointer.get());
}
// Setup
this->tiedconv_layer_->SetUp(up_top_vec_, &tiedconv_top_vec_);
if (Caffe::phase() == Caffe::TRAIN) {
for (int i = 0; i < up_top_vec_.size(); ++i) {
LOG(INFO) << " Top shape: " << tiedconv_top_vec_[i]->channels() << " "
<< tiedconv_top_vec_[i]->height() << " "
<< tiedconv_top_vec_[i]->width();
}
}
// Connect the W and b of tiedconv_layer to TI's blobs_ to make everything
// seem normal:
this->blobs_ = tiedconv_layer_->blobs_;
// DownPool
if (Caffe::phase() == Caffe::TRAIN)
LOG(INFO) << " Creating DownPooling Layer in "
<< this->layer_param_.name();
this->downpool_layer_ = new DownPoolingLayer<Dtype>(this->layer_param_);
// no need to make new blobs here bc downpool_bottom_vec_ == tiedconv_top_vec_
// and the top that net provides is the top for Downpool
this->downpool_layer_->SetUp(tiedconv_top_vec_, top);
}
/// @brief Append a new top blob to the net.
void faceRecognition::AppendTop(int layer_id)
{
std::shared_ptr<caffe::Blob<float> > blob_pointer(new caffe::Blob<float>());
blobs_.push_back(blob_pointer);
top_vecs_[layer_id].push_back(blob_pointer.get());
}