ssa_test.cpp

Go to the documentation of this file.

#include <cob_people_detection/subspace_analysis.h>
#include <cob_people_detection/face_normalizer.h>
#include <opencv2/opencv.hpp>
#include <iostream>
#include <fstream>
#include <sys/time.h>
#include <boost/timer.hpp>

bool preprocess(cv::Mat& img, cv::Mat& xyz, FaceNormalizer* fn, bool normalize, cv::Size& norm_size, cv::Mat& dm)
{
        //cv::Size norm_size=cv::Size(120,120);
        bool valid = true;
        if (normalize)
        {
                valid = fn->normalizeFace(img, xyz, norm_size, dm);

                // dm.convertTo(dm,CV_8UC1);
                // cv::equalizeHist(dm,dm);
                // cv::imshow("normalized",dm);
                // cv::waitKey(5);

        }
        else
        {
                cv::resize(img, img, norm_size);
        }

        dm.convertTo(dm, CV_64FC1);
        img.convertTo(img, CV_64FC1);
        return valid;
}
bool preprocess(cv::Mat& img, FaceNormalizer* fn, bool normalize, cv::Size& norm_size)
{
        bool valid = true;
        //cv::Size norm_size=cv::Size(120,120);
        if (normalize)
        {
                valid = fn->normalizeFace(img, norm_size);
                //cv::imshow("normalized",img);
                //cv::waitKey(5);

        }
        else
        {
                cv::resize(img, img, norm_size);
        }

        img.convertTo(img, CV_64FC1);
        return valid;
}

int main(int argc, const char *argv[])
{

        FaceNormalizer::FNConfig config;
        config.eq_ill = true;
        config.align = false;
        config.resize = true;
        config.cvt2gray = true;
        config.extreme_illumination_condtions = false;

        FaceNormalizer* fn = new FaceNormalizer();
        fn->init(config);

        // parse input arguments from command line
        std::string method_str, classifier_str;
        bool use_xyz = true;
        bool normalizer = false;
        if (argc == 1)
        {
                method_str = "FISHER";
                classifier_str = "SVM";
                normalizer = false;
        }

        else if (argc == 2)
        {
                method_str = argv[1];
                classifier_str = "KNN";
                normalizer = false;
        }

        else if (argc == 3)
        {
                method_str = argv[1];
                classifier_str = argv[2];
                normalizer = false;
        }

        else if (argc == 4)
        {
                method_str = argv[1];
                classifier_str = argv[2];

                if (std::strcmp(argv[3], "0") == 0)
                        normalizer = false;
                if (std::strcmp(argv[3], "1") == 0)
                        normalizer = true;

                if (std::strcmp(argv[4], "0") == 0)
                        use_xyz = false;
                if (std::strcmp(argv[4], "1") == 0)
                        use_xyz = true;
        }

        else if (argc == 5)
        {
                method_str = argv[1];
                classifier_str = argv[2];
                if (std::strcmp(argv[3], "0") == 0)
                        normalizer = false;
                if (std::strcmp(argv[3], "1") == 0)
                        normalizer = true;

                if (std::strcmp(argv[4], "0") == 0)
                        use_xyz = false;
                if (std::strcmp(argv[4], "1") == 0)
                        use_xyz = true;
        }

        //  Configure input params for subspace analysis

        SubspaceAnalysis::Method method;
        SubspaceAnalysis::Classifier classifier;

        if (!method_str.compare("FISHER"))
        {
                std::cout << "FISHER" << std::endl;
                method = SubspaceAnalysis::METH_FISHER;
        }
        else if (!method_str.compare("EIGEN"))
        {
                std::cout << "EIGEN" << std::endl;
                method = SubspaceAnalysis::METH_EIGEN;
        }
        else if (!method_str.compare("LDA2D"))
        {
                std::cout << "LDA2D" << std::endl;
                method = SubspaceAnalysis::METH_LDA2D;
        }
        else if (!method_str.compare("PCA2D"))
        {
                std::cout << "PCA2D" << std::endl;
                method = SubspaceAnalysis::METH_PCA2D;
        }
        else
        {
                std::cout << "ERROR: invalid method - use FISHER or EIGEN" << std::endl;
        }

        if (!classifier_str.compare("KNN"))
        {
                std::cout << "KNN" << std::endl;
                classifier = SubspaceAnalysis::CLASS_KNN;
        }
        else if (!classifier_str.compare("DIFFS"))
        {
                std::cout << "DIFFS" << std::endl;
                classifier = SubspaceAnalysis::CLASS_DIFS;
        }
        else if (!classifier_str.compare("SVM"))
        {
                std::cout << "SVM" << std::endl;
                classifier = SubspaceAnalysis::CLASS_SVM;
        }
        else if (!classifier_str.compare("RF"))
        {
                std::cout << "RF" << std::endl;
                classifier = SubspaceAnalysis::CLASS_RF;
        }
        else
        {
                std::cout << "ERROR: invalid classifier - use KNN or DIFFS or SVM" << std::endl;
        }

        std::cout << "SSA test configuration:" << std::endl;
        std::cout << "classifier: " << classifier_str << std::endl;
        std::cout << "method: " << method_str << std::endl;
        std::cout << "normalizing: " << normalizer << std::endl;
        std::cout << "use xyz: " << use_xyz << std::endl;
        //HOME
        //std::string training_set_path=     "/home/tom/git/care-o-bot/cob_people_perception/cob_people_detection/debug/eval/eval_tool_files/training_set_list";
        //std::string training_set_xyz_path= "/home/tom/git/care-o-bot/cob_people_perception/cob_people_detection/debug/eval/eval_tool_files/training_set_xyz_list";
        //std::string probe_file_path=       "/home/tom/git/care-o-bot/cob_people_perception/cob_people_detection/debug/eval/eval_tool_files/probe_file_list";
        //std::string probe_file_xyz_path=   "/home/tom/git/care-o-bot/cob_people_perception/cob_people_detection/debug/eval/eval_tool_files/probe_file_xyz_list";
        //IPA
        std::string training_set_path = "/share/goa-tz/people_detection/eval/eval_tool_files/training_set_list";
        std::string training_set_xyz_path = "/share/goa-tz/people_detection/eval/eval_tool_files/training_set_xyz_list";
        std::string probe_file_path = "/share/goa-tz/people_detection/eval/eval_tool_files/probe_file_list";
        std::string probe_file_xyz_path = "/share/goa-tz/people_detection/eval/eval_tool_files/probe_file_xyz_list";

        //read probe file
        std::ifstream probe_file_stream(probe_file_path.c_str());

        std::string probe_file;
        std::vector < std::string > probe_file_vec;

        while (probe_file_stream >> probe_file)
        {
                //std::cout<<probe_file<<std::endl;
                probe_file_vec.push_back(probe_file);

        }

        std::vector < std::string > probe_file_xyz_vec;
        if (use_xyz)
        {
                std::ifstream probe_file_xyz_stream(probe_file_xyz_path.c_str());
                std::string probe_file_xyz;

                while (probe_file_xyz_stream >> probe_file_xyz)
                {
                        //std::cout<<probe_file<<std::endl;
                        probe_file_xyz_vec.push_back(probe_file_xyz);

                }
        }

        // read training set
        std::ifstream in_file(training_set_path.c_str());
        std::string img_file;

        int label = 0;
        std::vector < std::string > in_vec;
        std::vector<int> label_vec;

        while (in_file >> img_file)
        {
                if (std::strcmp(img_file.c_str(), "$$") == 0)
                {

                        label++;
                }
                else
                {
                        in_vec.push_back(img_file);
                        label_vec.push_back(label);
                }

        }

        std::vector < std::string > in_vec_xyz;
        if (use_xyz)
        {
                // read xyz data
                std::ifstream in_file_xyz(training_set_xyz_path.c_str());
                std::string xml_file;

                while (in_file_xyz >> xml_file)
                {
                        if (!std::strcmp(xml_file.c_str(), "$$") == 0)
                        {
                                in_vec_xyz.push_back(xml_file);
                        }

                }
        }

        if ((use_xyz == true) && (in_vec.size() != in_vec_xyz.size() || probe_file_vec.size() != probe_file_xyz_vec.size()))
        {
                use_xyz = false;
                std::cerr << "Error - not for every image 3d information could be loaded - ignoring 3d information\n";
        }

        int num_classes = label;
        cv::Size norm_size;
        double aspect_ratio = 1;
        // load training images
        std::vector<cv::Mat> img_vec;
        std::vector<cv::Mat> dm_vec;

        std::string invalid_path = "/share/goa-tz/people_detection/eval/eval_tool_files/nrm_failed";
        //std::string invalid_path="/home/tom/git/care-o-bot/cob_people_perception/cob_people_detection/eval/eval_tool_files/nrm_failed";
        std::ofstream os_inv(invalid_path.c_str());
        bool valid;
        for (int i = 0; i < in_vec.size(); i++)
        {
                cv::Mat img;
                cv::Mat xyz;
                if (use_xyz)
                {
                        cv::FileStorage fs(in_vec_xyz[i], FileStorage::READ);
                        fs["depth"] >> xyz;
                        fs["color"] >> img;
                        fs.release();
                }
                else
                {
                        img = cv::imread(in_vec[i], 0);
                }

                if (i == 0)
                {
                        aspect_ratio = double(img.cols) / double(img.rows);
                        norm_size = cv::Size(round(160 * aspect_ratio), 160);
                        //norm_size=cv::Size(img.rows,img.cols);
                }
                valid = true;
                cv::Mat dm;
                if (use_xyz)
                        valid = preprocess(img, xyz, fn, normalizer, norm_size, dm);
                //if(use_xyz)valid=preprocess(img,fn,normalizer,norm_size);
                if (!use_xyz)
                        valid = preprocess(img, fn, normalizer, norm_size);

                img_vec.push_back(img);
                if (use_xyz)
                        dm_vec.push_back(dm);

                if (!valid)
                {
                        os_inv << in_vec[i] << "\n";
                }

        }

        // load test images
        std::vector<cv::Mat> probe_mat_vec;
        std::vector<cv::Mat> probe_dm_vec;
        for (int i = 0; i < probe_file_vec.size(); i++)
        {
                std::stringstream ostr, nstr;
                nstr << "/share/goa-tz/people_detection/eval/picdump/";
                //nstr<<"/home/tom/git/care-o-bot/cob_people_perception/cob_people_detection/debug/eval/picdump/";
                ostr << nstr.str().c_str() << i << "_orig" << ".jpg";

                cv::Mat probe_xyz, probe_img;
                if (use_xyz)
                {
                        cv::FileStorage fs(probe_file_xyz_vec[i], FileStorage::READ);
                        fs["depth"] >> probe_xyz;
                        fs["color"] >> probe_img;
                        fs.release();
                }
                else
                {
                        probe_img = cv::imread(probe_file_vec[i], 0);
                }

                cv::imwrite(ostr.str().c_str(), probe_img);

                valid = true;
                cv::Mat dm;
                if (use_xyz)
                        valid = preprocess(probe_img, probe_xyz, fn, normalizer, norm_size, dm);
                if (!use_xyz)
                        valid = preprocess(probe_img, fn, normalizer, norm_size);

                cv::Mat oimg;
                probe_img.convertTo(oimg, CV_8UC1);
                //cv::equalizeHist(oimg,oimg);
                nstr << i << "_norm" << ".jpg";
                cv::imwrite(nstr.str().c_str(), oimg);

                probe_mat_vec.push_back(probe_img);
                if (use_xyz)
                        probe_dm_vec.push_back(dm);

                if (!valid)
                {
                        os_inv << probe_file_vec[i] << "\n";
                }

        }
        os_inv.close();

        std::cout << "Size Training Set= " << img_vec.size() << std::endl;
        std::cout << "Size Test Set= " << probe_file_vec.size() << std::endl;

        int ss_dim = num_classes;

        SubspaceAnalysis::FaceRecognizer* EFF = new SubspaceAnalysis::FaceRecognizer();
        // calculate Model

        // timeval t1,t2,t3,t4;
        // gettimeofday(&t1,NULL);
        boost::timer t;
        EFF->trainModel(img_vec, label_vec, ss_dim, method, true, false);
        //gettimeofday(&t2,NULL);jj

        //open output file
        std::string path = "/share/goa-tz/people_detection/eval/eval_tool_files/classification_labels";
        std::string probabilities_path = "/share/goa-tz/people_detection/eval/eval_tool_files/classification_probabilities";
        std::string timing_path = "/share/goa-tz/people_detection/eval/eval_tool_files/timing";
        //std::string path = "/home/tom/git/care-o-bot/cob_people_perception/cob_people_detection/debug/eval/eval_tool_files/classified_output";
        std::ofstream os(path.c_str());
        std::ofstream probabilities_os(probabilities_path.c_str());
        std::ofstream timing_os(timing_path.c_str(), std::ofstream::app);

        timing_os << t.elapsed() << ",";
        std::cout << ">>>>>>>>>>training time = " << t.elapsed() << std::endl;

        std::cout << "EFF model computed" << std::endl;
        //EFF->loadModelFromFile("/share/goa-tz/people_detection/debug/rdata.xml",true);


        SubspaceAnalysis::FaceRecognizer* EFF_depth = new SubspaceAnalysis::FaceRecognizer();
        //if(use_xyz)
        //{
        //EFF_depth->trainModel(dm_vec,label_vec,ss_dim,method,true,false);
        //}


        //restart timer
        t.restart();
        for (int i = 0; i < probe_mat_vec.size(); i++)
        {
                cv::Mat probe = probe_mat_vec[i];
                int c_EFF;
                cv::Mat coeff_EFF;
                double DFFS_EFF;
                cv::Mat probabilities;
                if (method == SubspaceAnalysis::METH_LDA2D || method == SubspaceAnalysis::METH_PCA2D)
                {
                        std::vector<cv::Mat> coeff_EFF_vec;
                        EFF->projectToSubspace2D(probe, coeff_EFF_vec, DFFS_EFF);
                        EFF->classify(coeff_EFF_vec[0], classifier, c_EFF, probabilities);
                }
                else
                {
                        EFF->projectToSubspace(probe, coeff_EFF, DFFS_EFF);
                        EFF->classify(coeff_EFF, classifier, c_EFF, probabilities);
                }

                //c_EFF=model->predict(probe);
                //std::cout<<"RGB CLASS"<<c_EFF<<std::endl;

                //For use with depth data
                int c_EFF_dm;
                cv::Mat coeff_EFF_dm;
                double DFFS_EFF_dm;
                //if(use_xyz)
                //{
                cv::Mat probe_dm = probe_dm_vec[i];
                //EFF_depth->projectToSubspace(probe_dm,coeff_EFF_dm,DFFS_EFF_dm);
                //EFF_depth->classify(coeff_EFF_dm,classifier,c_EFF_dm);

                //}

                //Output to classified file
                os << c_EFF << "\n";
                probabilities_os << probabilities << "\n";
        }
        std::cout << ">>>>>>>>>>recognition time = " << t.elapsed() << std::endl;
        timing_os << t.elapsed() << std::endl;

        os.close();
        std::cout << "EFF classified" << std::endl;

        cv::Mat m1_evec, m1_eval, m1_avg, m1_pmd;

        //EFF->getModel(m1_evec,m1_eval,m1_avg,m1_pmd);
        //EFF->saveModel("/share/goa-tz/people_detection/debug/test.xml");


        //SubspaceAnalysis::FaceRecognizer* m2=new SubspaceAnalysis::FaceRecognizer();


        //m2->loadModel(m1_evec,m1_eval,m1_avg,m1_pmd,label_vec,false);
        //m2->loadModelFromFile("/share/goa-tz/people_detection/debug/rdata.xml",true);

        //double m2_dffs;
        //cv::Mat m2_coeff;
        //int m2_c;
        //cv::Mat probe=probe_mat_vec[0];
        //m2->projectToSubspace(probe,m2_coeff,m2_dffs);
        //m2->classify(m2_coeff,classifier,m2_c);


        // The following line predicts the label of a given
        // test image:
        //int predictedLabel = model->predict(testSample);


        return 0;
}