数据层:Image_data_layer层主要用于caffe来自图片的数据处理。
层类型:ImageData
必须设置的参数:①source:一个文本文件的名字,每行给定一个图片文件的名称和标签;②batch_size:每一次处理的数据个数,即图片数。
可选参数:①rand_skip:在开始时,跳过某个图片的输入;②shuffle:随机打乱顺序;③new_height,new_width:如果设置,则对图片进行Resize
(1)caffe.proto层参数定义
message ImageDataParameter {
optional string source = 1;//数据文件
optional uint32 batch_size = 4 [default = 1];
optional uint32 rand_skip = 7 [default = 0];//随机跳过
optional bool shuffle = 8 [default = false];//随机打乱
optional uint32 new_height = 9 [default = 0];//高度
optional uint32 new_width = 10 [default = 0];//宽度
optional bool is_color = 11 [default = true];//彩色图片
optional float scale = 2 [default = 1];
optional string mean_file = 3;//均值文件
optional uint32 crop_size = 5 [default = 0];//裁剪尺寸
optional bool mirror = 6 [default = false];//镜像翻转
optional string root_folder = 12 [default = ""];//图片存放根目录
}
(2)DataLayerSetUp层创建
template <typename Dtype>
void ImageDataLayer<Dtype>::DataLayerSetUp(const vector<Blob<Dtype>*>& bottom,
const vector<Blob<Dtype>*>& top) {
//根据层参数定义文件获取层参数:new_height、new_width、is_color、root_folder
const int new_height = this->layer_param_.image_data_param().new_height();//图像宽度
const int new_width = this->layer_param_.image_data_param().new_width();//图像高度
const bool is_color = this->layer_param_.image_data_param().is_color();//是否彩色图像
string root_folder = this->layer_param_.image_data_param().root_folder();//图像目录
//只支持图像宽度和高度同样大小的图片
CHECK((new_height == 0 && new_width == 0) ||
(new_height > 0 && new_width > 0)) << "Current implementation requires "
"new_height and new_width to be set at the same time.";
// Read the file with filenames and labels
//读取文件路径source(列表文件)和文件内容信息:文件名+标签
const string& source = this->layer_param_.image_data_param().source();
LOG(INFO) << "Opening file " << source;
std::ifstream infile(source.c_str());
string line;
size_t pos;
int label;
while (std::getline(infile, line)) {
pos = line.find_last_of(' ');//查找最后一个空格
label = atoi(line.substr(pos + 1).c_str());//将最后一个代表label信息的字符转化为对应int型
//lines_存放图片的位置(pos)和标签信息(label)
lines_.push_back(std::make_pair(line.substr(0, pos), label));
}
CHECK(!lines_.empty()) << "File is empty";
//是否需要打乱文件顺序
if (this->layer_param_.image_data_param().shuffle()) {
//若指定shuffle参数,则对图片进行随机打乱
// randomly shuffle data
LOG(INFO) << "Shuffling data";
const unsigned int prefetch_rng_seed = caffe_rng_rand();
prefetch_rng_.reset(new Caffe::RNG(prefetch_rng_seed));
ShuffleImages();
} else {
if (this->phase_ == TRAIN && Caffe::solver_rank() > 0 &&
this->layer_param_.image_data_param().rand_skip() == 0) {
LOG(WARNING) << "Shuffling or skipping recommended for multi-GPU";
}
}
LOG(INFO) << "A total of " << lines_.size() << " images.";
lines_id_ = 0;
// Check if we would need to randomly skip a few data points
if (this->layer_param_.image_data_param().rand_skip()) {
unsigned int skip = caffe_rng_rand() %//产生随机数
this->layer_param_.image_data_param().rand_skip();
LOG(INFO) << "Skipping first " << skip << " data points.";
CHECK_GT(lines_.size(), skip) << "Not enough points to skip";
lines_id_ = skip;
}
// Read an image, and use it to initialize the top blob.
//读取图像文件:cv_image
cv::Mat cv_img = ReadImageToCVMat(root_folder + lines_[lines_id_].first,
new_height, new_width, is_color);
CHECK(cv_img.data) << "Could not load " << lines_[lines_id_].first;
// Use data_transformer to infer the expected blob shape from a cv_image.
//根据cv_image推断Blob的shape
vector<int> top_shape = this->data_transformer_->InferBlobShape(cv_img);
//设置transformed_data_的形状
this->transformed_data_.Reshape(top_shape);
// Reshape prefetch_data and top[0] according to the batch_size.
//设置batch_size
const int batch_size = this->layer_param_.image_data_param().batch_size();
CHECK_GT(batch_size, 0) << "Positive batch size required";//batch_size>0
top_shape[0] = batch_size;
//设置预取数组中数据data_的形状
for (int i = 0; i < this->prefetch_.size(); ++i) {
this->prefetch_[i]->data_.Reshape(top_shape);
}
//设置输出数据top[0]的形状
top[0]->Reshape(top_shape);
LOG(INFO) << "output data size: " << top[0]->num() << ","
<< top[0]->channels() << "," << top[0]->height() << ","
<< top[0]->width();
// label
//设置输出标签top[1]的形状
vector<int> label_shape(1, batch_size);
top[1]->Reshape(label_shape);
//设置预取数组中标签label_的形状
for (int i = 0; i < this->prefetch_.size(); ++i) {
this->prefetch_[i]->label_.Reshape(label_shape);
}
}
(3)ShuffleImage随机打乱
template <typename Dtype>
void ImageDataLayer<Dtype>::ShuffleImages() {
caffe::rng_t* prefetch_rng =
//产生打乱图像顺序的数组prefetch_rng
static_cast<caffe::rng_t*>(prefetch_rng_->generator());
//打乱lines顺序,每个lines对应存放一个图像的信息,即将图像顺序打乱
shuffle(lines_.begin(), lines_.end(), prefetch_rng);
}
(4)load_batch批处理
template <typename Dtype>
void ImageDataLayer<Dtype>::load_batch(Batch<Dtype>* batch) {
CPUTimer batch_timer;
batch_timer.Start();
double read_time = 0;
double trans_time = 0;
CPUTimer timer;
CHECK(batch->data_.count());
CHECK(this->transformed_data_.count());
//根据参数定义文件获取层参数:batch_size、new_height、new_width、is_color、root_folder
ImageDataParameter image_data_param = this->layer_param_.image_data_param();
const int batch_size = image_data_param.batch_size();
const int new_height = image_data_param.new_height();
const int new_width = image_data_param.new_width();
const bool is_color = image_data_param.is_color();
string root_folder = image_data_param.root_folder();
// Reshape according to the first image of each batch
// on single input batches allows for inputs of varying dimension.
cv::Mat cv_img = ReadImageToCVMat(root_folder + lines_[lines_id_].first,
new_height, new_width, is_color);//读取图像
CHECK(cv_img.data) << "Could not load " << lines_[lines_id_].first;
// Use data_transformer to infer the expected blob shape from a cv_img.
vector<int> top_shape = this->data_transformer_->InferBlobShape(cv_img);//推断图像形状
this->transformed_data_.Reshape(top_shape);//设置transformed_data_的形状
// Reshape batch according to the batch_size.
//设置batch_size
top_shape[0] = batch_size;
batch->data_.Reshape(top_shape);
//预取数据data和标签label
Dtype* prefetch_data = batch->data_.mutable_cpu_data();
Dtype* prefetch_label = batch->label_.mutable_cpu_data();
// datum scales
//读取一批图像进行预处理
const int lines_size = lines_.size();
for (int item_id = 0; item_id < batch_size; ++item_id) {
// get a blob
timer.Start();
CHECK_GT(lines_size, lines_id_);
cv::Mat cv_img = ReadImageToCVMat(root_folder + lines_[lines_id_].first,
new_height, new_width, is_color);
//加载第一张图像
CHECK(cv_img.data) << "Could not load " << lines_[lines_id_].first;
read_time += timer.MicroSeconds();
timer.Start();
// Apply transformations (mirror, crop...) to the image
int offset = batch->data_.offset(item_id);
this->transformed_data_.set_cpu_data(prefetch_data + offset);
this->data_transformer_->Transform(cv_img, &(this->transformed_data_));
trans_time += timer.MicroSeconds();//统计预处理时间
prefetch_label[item_id] = lines_[lines_id_].second;
// go to the next iter
lines_id_++;
if (lines_id_ >= lines_size) {//是否是图像目录中的最后一张图像
// We have reached the end. Restart from the first.
DLOG(INFO) << "Restarting data prefetching from start.";
lines_id_ = 0;
if (this->layer_param_.image_data_param().shuffle()) {
ShuffleImages();//打乱图像索引的顺序
}
}
}
batch_timer.Stop();
DLOG(INFO) << "Prefetch batch: " << batch_timer.MilliSeconds() << " ms.";
DLOG(INFO) << " Read time: " << read_time / 1000 << " ms.";
DLOG(INFO) << "Transform time: " << trans_time / 1000 << " ms.";
}
(5)Usage使用
用bottom来输入数据,top来输出数据。在数据层至少有一个命名为data的top,如果有第二个top,一般命名为label。这种(data,label)的配对是分类模型所必须的。
name: "CaffeNet"
layer {
name: "data"//层名称,可随意取
type: "ImageData"//层类型
top: "data"
top: "label"//只有输出top,没有输入bottom
transform_param {//数据预处理
//是否开启镜像
mirror: false
//裁剪一个227*227的图像块
crop_size: 227
//用一个配置文件进行去均值操作
mean_file: "data/ilsvrc12/imagenet_mean.binaryproto"
}
image_data_param {
source: "examples/_temp/file_list.txt"
batch_size: 50
new_height: 256
new_width: 256
}
}
本文主要参考博文链接。如有错误,欢迎指正,谢谢!