netUtils.h

#ifndef MTF_NET_UTILS_H
#define MTF_NET_UTILS_H

#include <caffe/layers/memory_data_layer.hpp>
#include "mtf/Macros/common.h"
#include <caffe/caffe.hpp>
#include <vector>
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <iostream>
#include <iostream>
#include "boost/scoped_ptr.hpp"
#include "caffe/util/db.hpp"
#include "caffe/util/io.hpp"
#include "mtf/SM/RegNetParams.h"

using boost::scoped_ptr;
using namespace caffe;

_MTF_BEGIN_NAMESPACE

namespace utils{
    class MTFNet{
    public:
        MTFNet(RegNetParams _rgparams);

        boost::shared_ptr<caffe::Net<float> > net_; // Network object
        boost::shared_ptr<caffe::Solver<float > > solver_;
        cv::Size input_geometry_; // W*H Input Geometry to the network
        int num_channels_; //Number of Channels
        cv::Scalar channel_mean_;

        RegNetParams rgparams;

    private:
        void getGPUs(vector<int>* gpus);

    public:
        void preprocessBatch(std::vector<cv::Mat>& batch);

        void extractBatch(int bs, int bi, std::vector<cv::Mat> t_data, std::vector<cv::Mat> t_labels, std::vector<cv::Mat> &t_data_b, std::vector<cv::Mat> &t_labels_b);
    };

    MTFNet *createNetwork(RegNetParams rgparams, char *conv_model = nullptr, char *last_conv_layer = nullptr,
        int n_layers = 2, int *n_neurons = nullptr, char *activ_fn = nullptr);

    float *forwardPass(MTFNet *network, cv::Mat patch);

    void train(MTFNet *network, std::vector<cv::Mat> training_data, std::vector<cv::Mat> training_labels);

    void testingInputBlobs(boost::shared_ptr<caffe::Net<float> > net_);
}

_MTF_END_NAMESPACE
#endif