Task02 Денис Сарахатунов СПбГУ

src/phg/matching/descriptor_matcher.cpp

@@ -1,14 +1,23 @@
 #include "descriptor_matcher.h"
 
 #include <opencv2/flann/miniflann.hpp>
+#include <algorithm>
+#include <vector>
 #include "flann_factory.h"
 
 void phg::DescriptorMatcher::filterMatchesRatioTest(const std::vector<std::vector<cv::DMatch>> &matches,
                                                     std::vector<cv::DMatch> &filtered_matches)
 {
     filtered_matches.clear();
+    filtered_matches.reserve(matches.size());
 
-    throw std::runtime_error("not implemented yet");
+    const float ratio = 0.65f;
+    for (const std::vector<cv::DMatch> &knn : matches) {
+        if (knn.size() < 2)
+            continue;
+        if (knn[0].distance < ratio*knn[1].distance)
+            filtered_matches.push_back(knn[0]);
+    }
 }
 
 
@@ -35,42 +44,64 @@ void phg::DescriptorMatcher::filterMatchesClusters(const std::vector<cv::DMatch>
         points_query.at<cv::Point2f>(i) = keypoints_query[matches[i].queryIdx].pt;
         points_train.at<cv::Point2f>(i) = keypoints_train[matches[i].trainIdx].pt;
     }
-//
-//    // размерность всего 2, так что точное KD-дерево
-//    std::shared_ptr<cv::flann::IndexParams> index_params = flannKdTreeIndexParams(TODO);
-//    std::shared_ptr<cv::flann::SearchParams> search_params = flannKsTreeSearchParams(TODO);
-//
-//    std::shared_ptr<cv::flann::Index> index_query = flannKdTreeIndex(points_query, index_params);
-//    std::shared_ptr<cv::flann::Index> index_train = flannKdTreeIndex(points_train, index_params);
-//
-//    // для каждой точки найти total neighbors ближайших соседей
-//    cv::Mat indices_query(n_matches, total_neighbours, CV_32SC1);
-//    cv::Mat distances2_query(n_matches, total_neighbours, CV_32FC1);
-//    cv::Mat indices_train(n_matches, total_neighbours, CV_32SC1);
-//    cv::Mat distances2_train(n_matches, total_neighbours, CV_32FC1);
-//
-//    index_query->knnSearch(points_query, indices_query, distances2_query, total_neighbours, *search_params);
-//    index_train->knnSearch(points_train, indices_train, distances2_train, total_neighbours, *search_params);
-//
-//    // оценить радиус поиска для каждой картинки
-//    // NB: radius2_query, radius2_train: квадраты радиуса!
-//    float radius2_query, radius2_train;
-//    {
-//        std::vector<double> max_dists2_query(n_matches);
-//        std::vector<double> max_dists2_train(n_matches);
-//        for (int i = 0; i < n_matches; ++i) {
-//            max_dists2_query[i] = distances2_query.at<float>(i, total_neighbours - 1);
-//            max_dists2_train[i] = distances2_train.at<float>(i, total_neighbours - 1);
-//        }
-//
-//        int median_pos = n_matches / 2;
-//        std::nth_element(max_dists2_query.begin(), max_dists2_query.begin() + median_pos, max_dists2_query.end());
-//        std::nth_element(max_dists2_train.begin(), max_dists2_train.begin() + median_pos, max_dists2_train.end());
-//
-//        radius2_query = max_dists2_query[median_pos] * radius_limit_scale * radius_limit_scale;
-//        radius2_train = max_dists2_train[median_pos] * radius_limit_scale * radius_limit_scale;
-//    }
-//
-//    метч остается, если левое и правое множества первых total_neighbors соседей в радиусах поиска(radius2_query, radius2_train) имеют как минимум consistent_matches общих элементов
-//    // TODO заполнить filtered_matches
+
+    // размерность всего 2, так что точное KD-дерево
+    std::shared_ptr<cv::flann::IndexParams> index_params = flannKdTreeIndexParams(1);
+    std::shared_ptr<cv::flann::SearchParams> search_params = flannKsTreeSearchParams(std::max(n_matches, 64));
+
+    std::shared_ptr<cv::flann::Index> index_query = flannKdTreeIndex(points_query, index_params);
+    std::shared_ptr<cv::flann::Index> index_train = flannKdTreeIndex(points_train, index_params);
+
+    // для каждой точки найти total neighbors ближайших соседей
+    cv::Mat indices_query(n_matches, (int)total_neighbours, CV_32SC1);
+    cv::Mat distances2_query(n_matches, (int)total_neighbours, CV_32FC1);
+    cv::Mat indices_train(n_matches, (int)total_neighbours, CV_32SC1);
+    cv::Mat distances2_train(n_matches, (int)total_neighbours, CV_32FC1);
+
+    index_query->knnSearch(points_query, indices_query, distances2_query, (int)total_neighbours, *search_params);
+    index_train->knnSearch(points_train, indices_train, distances2_train, (int)total_neighbours, *search_params);
+
+    // оценить радиус поиска для каждой картинки
+    // NB: radius2_query, radius2_train: квадраты радиуса!
+    float radius2_query, radius2_train;
+    {
+        std::vector<double> max_dists2_query(n_matches);
+        std::vector<double> max_dists2_train(n_matches);
+        for (int i = 0; i < n_matches; ++i) {
+            max_dists2_query[i] = distances2_query.at<float>(i, (int)total_neighbours - 1);
+            max_dists2_train[i] = distances2_train.at<float>(i, (int)total_neighbours - 1);
+        }
+
+        int median_pos = n_matches / 2;
+        std::nth_element(max_dists2_query.begin(), max_dists2_query.begin() + median_pos, max_dists2_query.end());
+        std::nth_element(max_dists2_train.begin(), max_dists2_train.begin() + median_pos, max_dists2_train.end());
+
+        radius2_query = max_dists2_query[median_pos] * radius_limit_scale * radius_limit_scale;
+        radius2_train = max_dists2_train[median_pos] * radius_limit_scale * radius_limit_scale;
+    }
+
+    //    метч остается, если левое и правое множества первых total_neighbors соседей в радиусах поиска(radius2_query, radius2_train) имеют как минимум consistent_matches общих элементов
+    //    // TODO заполнить filtered_matches
+    std::vector<int> tmp(n_matches, -1);
+    filtered_matches.reserve(n_matches);
+
+    for (int i = 0; i < n_matches; ++i) {
+        for (int j = 0; j < total_neighbours; ++j) {
+            int idx_q = indices_query.at<int>(i, (int)j);
+            float dist2_q = distances2_query.at<float>(i, (int)j);
+            if (idx_q >= 0 && idx_q < n_matches && dist2_q <= radius2_query)
+                tmp[idx_q] = i;
+        }
+
+        int n_common = 0;
+        for (int j = 0; j < total_neighbours; ++j) {
+            int idx_t = indices_train.at<int>(i, (int)j);
+            float dist2_t = distances2_train.at<float>(i, (int)j);
+            if (idx_t >= 0 && idx_t < n_matches && dist2_t <= radius2_train && tmp[idx_t] == i)
+                ++n_common;
+        }
+
+        if (n_common >= (int)consistent_matches)
+            filtered_matches.push_back(matches[i]);
+    }
 }

src/phg/matching/flann_matcher.cpp

@@ -1,13 +1,14 @@
 #include <iostream>
+#include <cmath>
 #include "flann_matcher.h"
 #include "flann_factory.h"
 
 
 phg::FlannMatcher::FlannMatcher()
 {
     // параметры для приближенного поиска
-//    index_params = flannKdTreeIndexParams(TODO);
-//    search_params = flannKsTreeSearchParams(TODO);
+    index_params = flannKdTreeIndexParams(4);
+    search_params = flannKsTreeSearchParams(32);
 }
 
 void phg::FlannMatcher::train(const cv::Mat &train_desc)
@@ -17,5 +18,24 @@ void phg::FlannMatcher::train(const cv::Mat &train_desc)
 
 void phg::FlannMatcher::knnMatch(const cv::Mat &query_desc, std::vector<std::vector<cv::DMatch>> &matches, int k) const
 {
-    throw std::runtime_error("not implemented yet");
+    if (!flann_index)
+        throw std::runtime_error("not trained");
+    if (query_desc.empty()) {
+        matches.clear();
+        return;
+    }
+
+    cv::Mat indexes(query_desc.rows, k, CV_32SC1);
+    cv::Mat distances2(query_desc.rows, k, CV_32FC1);
+    flann_index->knnSearch(query_desc, indexes, distances2, k, *search_params);
+
+    matches.assign(query_desc.rows, {});
+    for (int qi = 0; qi < query_desc.rows; ++qi) {
+        matches[qi].reserve(k);
+        for (int ki = 0; ki < k; ++ki) {
+            int train_idx = indexes.at<int>(qi, ki);
+            float dist = std::sqrt(distances2.at<float>(qi, ki));
+            matches[qi].emplace_back(qi, train_idx, 0, dist);
+        }
+    }
 }

src/phg/sfm/homography.cpp

@@ -2,6 +2,8 @@
 
 #include <opencv2/calib3d/calib3d.hpp>
 #include <iostream>
+#include <cmath>
+#include <algorithm>
 
 namespace {
 
@@ -84,8 +86,8 @@ namespace {
             double w1 = ws1[i];
 
             // 8 elements of matrix + free term as needed by gauss routine
-//            A.push_back({TODO});
-//            A.push_back({TODO});
+            A.push_back({w1*x0, w1*y0, w1*w0, 0, 0, 0, -x1*x0, -x1*y0, x1*w0});
+            A.push_back({0, 0, 0, w1*x0, w1*y0, w1*w0, -y1*x0, -y1*y0, y1*w0});
         }
 
         int res = gauss(A, H);
@@ -168,57 +170,59 @@ namespace {
         // * (простое описание для понимания)
         // * [3] http://ikrisoft.blogspot.com/2015/01/ransac-with-contrario-approach.html
 
-//        const int n_matches = points_lhs.size();
-//
-//        // https://en.wikipedia.org/wiki/Random_sample_consensus#Parameters
-//        const int n_trials = TODO;
-//
-//        const int n_samples = TODO;
-//        uint64_t seed = 1;
-//        const double reprojection_error_threshold_px = 2;
-//
-//        int best_support = 0;
-//        cv::Mat best_H;
-//
-//        std::vector<int> sample;
-//        for (int i_trial = 0; i_trial < n_trials; ++i_trial) {
-//            randomSample(sample, n_matches, n_samples, &seed);
-//
-//            cv::Mat H = estimateHomography4Points(points_lhs[sample[0]], points_lhs[sample[1]], points_lhs[sample[2]], points_lhs[sample[3]],
-//                                                  points_rhs[sample[0]], points_rhs[sample[1]], points_rhs[sample[2]], points_rhs[sample[3]]);
-//
-//            int support = 0;
-//            for (int i_point = 0; i_point < n_matches; ++i_point) {
-//                try {
-//                    cv::Point2d proj = phg::transformPoint(points_lhs[i_point], H);
-//                    if (cv::norm(proj - cv::Point2d(points_rhs[i_point])) < reprojection_error_threshold_px) {
-//                        ++support;
-//                    }
-//                } catch (const std::exception &e)
-//                {
-//                    std::cerr << e.what() << std::endl;
-//                }
-//            }
-//
-//            if (support > best_support) {
-//                best_support = support;
-//                best_H = H;
-//
-//                std::cout << "estimateHomographyRANSAC : support: " << best_support << "/" << n_matches << std::endl;
-//
-//                if (best_support == n_matches) {
-//                    break;
-//                }
-//            }
-//        }
-//
-//        std::cout << "estimateHomographyRANSAC : best support: " << best_support << "/" << n_matches << std::endl;
-//
-//        if (best_support == 0) {
-//            throw std::runtime_error("estimateHomographyRANSAC : failed to estimate homography");
-//        }
-//
-//        return best_H;
+        const int n_matches = points_lhs.size();
+        if (n_matches < 4)
+            throw std::runtime_error("need more");
+
+        // https://en.wikipedia.org/wiki/Random_sample_consensus#Parameters
+        const int n_trials = 1000;
+
+        const int n_samples = 4;
+        uint64_t seed = 1;
+        const double reprojection_error_threshold_px = 2;
+
+        int best_support = 0;
+        cv::Mat best_H;
+
+        std::vector<int> sample;
+        for (int i_trial = 0; i_trial < n_trials; ++i_trial) {
+            randomSample(sample, n_matches, n_samples, &seed);
+
+            cv::Mat H = estimateHomography4Points(points_lhs[sample[0]], points_lhs[sample[1]], points_lhs[sample[2]], points_lhs[sample[3]],
+                                                    points_rhs[sample[0]], points_rhs[sample[1]], points_rhs[sample[2]], points_rhs[sample[3]]);
+
+            int support = 0;
+            for (int i_point = 0; i_point < n_matches; ++i_point) {
+                try {
+                    cv::Point2d proj = phg::transformPoint(points_lhs[i_point], H);
+                    if (cv::norm(proj - cv::Point2d(points_rhs[i_point])) < reprojection_error_threshold_px) {
+                        ++support;
+                    }
+                } catch (const std::exception &e)
+                {
+                    std::cerr << e.what() << std::endl;
+                }
+            }
+
+            if (support > best_support) {
+                best_support = support;
+                best_H = H;
+
+                std::cout << "estimateHomographyRANSAC : support: " << best_support << "/" << n_matches << std::endl;
+
+                if (best_support == n_matches) {
+                    break;
+                }
+            }
+        }
+
+        std::cout << "estimateHomographyRANSAC : best support: " << best_support << "/" << n_matches << std::endl;
+
+        if (best_support == 0) {
+            throw std::runtime_error("estimateHomographyRANSAC : failed to estimate homography");
+        }
+
+        return best_H;
     }
 
 }
@@ -238,7 +242,34 @@ cv::Mat phg::findHomographyCV(const std::vector<cv::Point2f> &points_lhs, const
 // таким преобразованием внутри занимается функции cv::perspectiveTransform и cv::warpPerspective
 cv::Point2d phg::transformPoint(const cv::Point2d &pt, const cv::Mat &T)
 {
-    throw std::runtime_error("not implemented yet");
+    if (T.rows != 3 || T.cols != 3) {
+        throw std::runtime_error("not 3x3 matrix");
+    }
+
+    auto x = pt.x;
+    auto y = pt.y;
+
+    double tx, ty, tw;
+    if (T.type() == CV_64FC1) {
+        tx = T.at<double>(0, 0)*x + T.at<double>(0, 1)*y + T.at<double>(0, 2);
+        ty = T.at<double>(1, 0)*x + T.at<double>(1, 1)*y + T.at<double>(1, 2);
+        tw = T.at<double>(2, 0)*x + T.at<double>(2, 1)*y + T.at<double>(2, 2);
+    } else if (T.type() == CV_32FC1) {
+        tx = T.at<float>(0, 0)*x + T.at<float>(0, 1)*y + T.at<float>(0, 2);
+        ty = T.at<float>(1, 0)*x + T.at<float>(1, 1)*y + T.at<float>(1, 2);
+        tw = T.at<float>(2, 0)*x + T.at<float>(2, 1)*y + T.at<float>(2, 2);
+    } else {
+        cv::Mat Td;
+        T.convertTo(Td, CV_64FC1);
+        tx = Td.at<double>(0, 0)*x + Td.at<double>(0, 1)*y + Td.at<double>(0, 2);
+        ty = Td.at<double>(1, 0)*x + Td.at<double>(1, 1)*y + Td.at<double>(1, 2);
+        tw = Td.at<double>(2, 0)*x + Td.at<double>(2, 1)*y + Td.at<double>(2, 2);
+    }
+
+    if (std::abs(tw) < 1e-12)
+        throw std::runtime_error("zero div");
+
+    return {tx/tw, ty/tw};
 }
 
 cv::Point2d phg::transformPointCV(const cv::Point2d &pt, const cv::Mat &T) {

src/phg/sfm/panorama_stitcher.cpp

@@ -23,7 +23,45 @@ cv::Mat phg::stitchPanorama(const std::vector<cv::Mat> &imgs,
     {
         // здесь надо посчитать вектор Hs
         // при этом можно обойтись n_images - 1 вызовами функтора homography_builder
-        throw std::runtime_error("not implemented yet");
+        std::vector<int> state(n_images, 0);
+        for (int i = 0; i < n_images; ++i) {
+            if (state[i] == 2)
+                continue;
+
+            std::vector<int> stack;
+            int v = i;
+            while (v >= 0) {
+                if (v >= n_images)
+                    throw std::runtime_error("invalid index");
+                if (state[v] == 2)
+                    break;
+                if (state[v] == 1)
+                    throw std::runtime_error("cycle");
+
+                state[v] = 1;
+                stack.push_back(v);
+                v = parent[v];
+            }
+
+            while (!stack.empty()) {
+                int node = stack.back();
+                stack.pop_back();
+
+                int p = parent[node];
+                if (p < 0) {
+                    Hs[node] = cv::Mat::eye(3, 3, CV_64FC1);
+                } else {
+                    cv::Mat H_i_to_p = homography_builder(imgs[node], imgs[p]);
+
+                    if (H_i_to_p.empty())
+                        throw std::runtime_error("empty");
+
+                    H_i_to_p.convertTo(H_i_to_p, CV_64FC1);
+                    Hs[node] = Hs[p] * H_i_to_p;
+                }
+                state[node] = 2;
+            }
+        }
     }
 
     bbox2<double, cv::Point2d> bbox;