internal/ceres/canonical_views_clustering_test.cc

   1 // Ceres Solver - A fast non-linear least squares minimizer
   2 // Copyright 2015 Google Inc. All rights reserved.
   3 // http://ceres-solver.org/
   4 //
   5 // Redistribution and use in source and binary forms, with or without
   6 // modification, are permitted provided that the following conditions are met:
   7 //
   8 // * Redistributions of source code must retain the above copyright notice,
   9 //   this list of conditions and the following disclaimer.
  10 // * Redistributions in binary form must reproduce the above copyright notice,
  11 //   this list of conditions and the following disclaimer in the documentation
  12 //   and/or other materials provided with the distribution.
  13 // * Neither the name of Google Inc. nor the names of its contributors may be
  14 //   used to endorse or promote products derived from this software without
  15 //   specific prior written permission.
  16 //
  17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  18 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20 // ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  21 // LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  22 // CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  23 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  24 // INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  25 // CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  26 // ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  27 // POSSIBILITY OF SUCH DAMAGE.
  28 //
  29 // Author: Sameer Agarwal (sameeragarwal@google.com)
  30 //         David Gallup (dgallup@google.com)
  31
  32 #include "ceres/canonical_views_clustering.h"
  33
  34 #include "ceres/collections_port.h"
  35 #include "ceres/graph.h"
  36 #include "gtest/gtest.h"
  37
  38 namespace ceres {
  39 namespace internal {
  40
  41 const int kVertexIds[] = {0, 1, 2, 3};
  42 class CanonicalViewsTest : public ::testing::Test {
  43  protected:
  44   virtual void SetUp() {
  45     // The graph structure is as follows.
  46     //
  47     // Vertex weights:   0      2      2      0
  48     //                   V0-----V1-----V2-----V3
  49     // Edge weights:        0.8    0.9    0.3
  50     const double kVertexWeights[] = {0.0, 2.0, 2.0, -1.0};
  51     for (int i = 0; i < 4; ++i) {
  52       graph_.AddVertex(i, kVertexWeights[i]);
  53     }
  54     // Create self edges.
  55     // CanonicalViews requires that every view "sees" itself.
  56     for (int i = 0; i < 4; ++i) {
  57       graph_.AddEdge(i, i, 1.0);
  58     }
  59
  60     // Create three edges.
  61     const double kEdgeWeights[] = {0.8, 0.9, 0.3};
  62     for (int i = 0; i < 3; ++i) {
  63       // The graph interface is directed, so remember to create both
  64       // edges.
  65       graph_.AddEdge(kVertexIds[i], kVertexIds[i + 1], kEdgeWeights[i]);
  66     }
  67   }
  68
  69   void ComputeClustering() {
  70     ComputeCanonicalViewsClustering(options_, graph_, &centers_, &membership_);
  71   }
  72
  73   WeightedGraph<int> graph_;
  74
  75   CanonicalViewsClusteringOptions options_;
  76   std::vector<int> centers_;
  77   HashMap<int, int> membership_;
  78 };
  79
  80 TEST_F(CanonicalViewsTest, ComputeCanonicalViewsTest) {
  81   options_.min_views = 0;
  82   options_.size_penalty_weight = 0.5;
  83   options_.similarity_penalty_weight = 0.0;
  84   options_.view_score_weight = 0.0;
  85   ComputeClustering();
  86
  87   // 2 canonical views.
  88   EXPECT_EQ(centers_.size(), 2);
  89   EXPECT_EQ(centers_[0], kVertexIds[1]);
  90   EXPECT_EQ(centers_[1], kVertexIds[3]);
  91
  92   // Check cluster membership.
  93   EXPECT_EQ(FindOrDie(membership_, kVertexIds[0]), 0);
  94   EXPECT_EQ(FindOrDie(membership_, kVertexIds[1]), 0);
  95   EXPECT_EQ(FindOrDie(membership_, kVertexIds[2]), 0);
  96   EXPECT_EQ(FindOrDie(membership_, kVertexIds[3]), 1);
  97 }
  98
  99 // Increases size penalty so the second canonical view won't be
 100 // chosen.
 101 TEST_F(CanonicalViewsTest, SizePenaltyTest) {
 102   options_.min_views = 0;
 103   options_.size_penalty_weight = 2.0;
 104   options_.similarity_penalty_weight = 0.0;
 105   options_.view_score_weight = 0.0;
 106   ComputeClustering();
 107
 108   // 1 canonical view.
 109   EXPECT_EQ(centers_.size(), 1);
 110   EXPECT_EQ(centers_[0], kVertexIds[1]);
 111 }
 112
 113
 114 // Increases view score weight so vertex 2 will be chosen.
 115 TEST_F(CanonicalViewsTest, ViewScoreTest) {
 116   options_.min_views = 0;
 117   options_.size_penalty_weight = 0.5;
 118   options_.similarity_penalty_weight = 0.0;
 119   options_.view_score_weight = 1.0;
 120   ComputeClustering();
 121
 122   // 2 canonical views.
 123   EXPECT_EQ(centers_.size(), 2);
 124   EXPECT_EQ(centers_[0], kVertexIds[1]);
 125   EXPECT_EQ(centers_[1], kVertexIds[2]);
 126 }
 127
 128 // Increases similarity penalty so vertex 2 won't be chosen despite
 129 // it's view score.
 130 TEST_F(CanonicalViewsTest, SimilarityPenaltyTest) {
 131   options_.min_views = 0;
 132   options_.size_penalty_weight = 0.5;
 133   options_.similarity_penalty_weight = 3.0;
 134   options_.view_score_weight = 1.0;
 135   ComputeClustering();
 136
 137   // 2 canonical views.
 138   EXPECT_EQ(centers_.size(), 1);
 139   EXPECT_EQ(centers_[0], kVertexIds[1]);
 140 }
 141
 142 }  // namespace internal
 143 }  // namespace ceres