caffe源码分析-DataTransformer

bear_fish

发布于 2019-02-25 11:53:25

5770

发布于 2019-02-25 11:53:25

文章被收录于专栏：用户2442861的专栏

本文主要分析caffe中DataTransformer这个类, 主要作用是：

将Datum类型或者cv::Mat，转化为caffe的Blob<Dtype>，并按照Transformation``Parameter参数对图像做处理，例如scale,mirro等
推断blob的shape

proto定义如下：

// Message that stores parameters used to apply transformation
// to the data layer's data
message TransformationParameter {

  optional float scale = 1 [default = 1];
  // Specify if we want to randomly mirror data.
  optional bool mirror = 2 [default = false];
  // Specify if we would like to randomly crop an image.
  optional uint32 crop_size = 3 [default = 0];
  // mean_file and mean_value cannot be specified at the same time
  optional string mean_file = 4;
  repeated float mean_value = 5;
  // Force the decoded image to have 3 color channels.
  optional bool force_color = 6 [default = false];
  // Force the decoded image to have 1 color channels.
  optional bool force_gray = 7 [default = false];
}

使用示例1：

layer {
  name: "mnist"
  type: "Data"
  top: "data"
  top: "label"
  include {
    phase: TEST
  }
  transform_param {
    scale: 0.00390625  # 1 / 255
  }
  data_param {
    source: "/home/xy/caffe-master/examples/mnist/mnist_test_lmdb"
    batch_size: 100
    backend: LMDB
  }
}

使用示例2：

transform_param {
  # randomly horizontally mirror the image
  mirror: 1
  crop_size: 227
  # substract mean value（RGB three channel）: or use mean.binaryproto file
  # mean_file: name_of_mean_file.binaryproto
  mean_value: 104
  mean_value: 117
  mean_value: 123
}

下面看其核心的数据成员以及函数定义：

template <typename Dtype>
class DataTransformer {
public:
    explicit DataTransformer(const TransformationParameter& param, Phase phase);

    void Transform(const Datum& datum, Blob<Dtype>* transformed_blob);
    void Transform(const cv::Mat& cv_img, Blob<Dtype>* transformed_blob);

    vector<int> InferBlobShape(const Datum& datum);
    vector<int> InferBlobShape(const cv::Mat& cv_img);

protected:
    // Tranformation parameters
    TransformationParameter param_;

    shared_ptr<Caffe::RNG> rng_;
    Phase phase_;
    Blob<Dtype> data_mean_;
    vector<Dtype> mean_values_;
};

下面仅仅给出将Datum类型转化为caffe的Blob， cv::Mat的转化同理.

template<typename Dtype>
void DataTransformer<Dtype>::Transform(const Datum& datum,
                                       Dtype* transformed_data) {
    const string& data = datum.data();
    const int datum_channels = datum.channels();
    const int datum_height = datum.height();
    const int datum_width = datum.width();

    const int crop_size = param_.crop_size();
    const Dtype scale = param_.scale();
    const bool do_mirror = param_.mirror() && Rand(2);
    const bool has_mean_file = param_.has_mean_file();
    const bool has_uint8 = data.size() > 0;
    const bool has_mean_values = mean_values_.size() > 0;

    Dtype* mean = NULL;
    if (has_mean_file) {
        mean = data_mean_.mutable_cpu_data();
    }
    if (has_mean_values) {
        if (datum_channels > 1 && mean_values_.size() == 1) {
            // Replicate the mean_value for simplicity
            for (int c = 1; c < datum_channels; ++c) {
                mean_values_.push_back(mean_values_[0]);
            }
        }
    }

    int height = datum_height;
    int width = datum_width;

    int h_off = 0;
    int w_off = 0;
    if (crop_size) {
        height = crop_size;
        width = crop_size;
        // We only do random crop when we do training.
        if (phase_ == TRAIN) {
            h_off = Rand(datum_height - crop_size + 1);
            w_off = Rand(datum_width - crop_size + 1);
        } else {
            h_off = (datum_height - crop_size) / 2;
            w_off = (datum_width - crop_size) / 2;
        }
    }

    Dtype datum_element;
    int top_index, data_index;
    for (int c = 0; c < datum_channels; ++c) {
        for (int h = 0; h < height; ++h) {
            for (int w = 0; w < width; ++w) {
                data_index = (c * datum_height + h_off + h) * datum_width + w_off + w;
                if (do_mirror) {
                    top_index = (c * height + h) * width + (width - 1 - w);
                } else {
                    top_index = (c * height + h) * width + w;
                }
                if (has_uint8) {
                    datum_element =
                            static_cast<Dtype>(static_cast<uint8_t>(data[data_index]));
                } else {
                    datum_element = datum.float_data(data_index);
                }
                if (has_mean_file) {
                    transformed_data[top_index] =
                            (datum_element - mean[data_index]) * scale;
                } else {
                    if (has_mean_values) {
                        transformed_data[top_index] =
                                (datum_element - mean_values_[c]) * scale;
                    } else {
                        transformed_data[top_index] = datum_element * scale;
                    }
                }
            }
        }
    }
}