Merge pull request #1616 from borglab/cherrypick-transitivity-fix-dsftrackgenerator

Cherrypick commits for release/4.2, that include transitivity fix for DsfTrackGenerator

gtsam/sfm/DsfTrackGenerator.cpp

@@ -20,6 +20,7 @@
 
 #include <algorithm>
 #include <iostream>
+#include <iomanip>
 
 namespace gtsam {
 
@@ -38,7 +39,8 @@ static DSFMapIndexPair generateDSF(const MatchIndicesMap& matches) {
     // Image pair is (i1,i2).
     size_t i1 = pair_indices.first;
     size_t i2 = pair_indices.second;
-    for (size_t k = 0; k < corr_indices.rows(); k++) {
+    size_t m = static_cast<size_t>(corr_indices.rows());
+    for (size_t k = 0; k < m; k++) {
       // Measurement indices are found in a single matrix row, as (k1,k2).
       size_t k1 = corr_indices(k, 0), k2 = corr_indices(k, 1);
       // Unique key for DSF is (i,k), representing keypoint index in an image.
@@ -128,7 +130,7 @@ std::vector<SfmTrack2d> tracksFromPairwiseMatches(
   }
 
   // TODO(johnwlambert): return the Transitivity failure percentage here.
-  return tracks2d;
+  return validTracks;
 }
 
 }  // namespace gtsfm

python/gtsam/tests/test_DsfTrackGenerator.py

@@ -4,18 +4,44 @@ Authors: John Lambert
 """
 
 import unittest
+from typing import Dict, Tuple
 
-import gtsam
 import numpy as np
-from gtsam import (IndexPair, KeypointsVector, MatchIndicesMap, Point2,
-                   SfmMeasurementVector, SfmTrack2d)
 from gtsam.gtsfm import Keypoints
 from gtsam.utils.test_case import GtsamTestCase
 
+import gtsam
+from gtsam import (IndexPair, KeypointsVector, MatchIndicesMap, Point2,
+                   SfmMeasurementVector, SfmTrack2d)
+
 
 class TestDsfTrackGenerator(GtsamTestCase):
     """Tests for DsfTrackGenerator."""
 
+    def test_generate_tracks_from_pairwise_matches_nontransitive(
+        self,
+    ) -> None:
+        """Tests DSF for non-transitive matches.
+
+        Test will result in no tracks since nontransitive tracks are naively
+        discarded by DSF.
+        """
+        keypoints = get_dummy_keypoints_list()
+        nontransitive_matches = get_nontransitive_matches()
+
+        # For each image pair (i1,i2), we provide a (K,2) matrix
+        # of corresponding keypoint indices (k1,k2).
+        matches = MatchIndicesMap()
+        for (i1, i2), correspondences in nontransitive_matches.items():
+            matches[IndexPair(i1, i2)] = correspondences
+
+        tracks = gtsam.gtsfm.tracksFromPairwiseMatches(
+            matches,
+            keypoints,
+            verbose=True,
+        )
+        self.assertEqual(len(tracks), 0, "Tracks not filtered correctly")
+
     def test_track_generation(self) -> None:
         """Ensures that DSF generates three tracks from measurements
         in 3 images (H=200,W=400)."""
@@ -23,20 +49,20 @@ class TestDsfTrackGenerator(GtsamTestCase):
         kps_i1 = Keypoints(np.array([[50.0, 60], [70, 80], [90, 100]]))
         kps_i2 = Keypoints(np.array([[110.0, 120], [130, 140]]))
 
-        keypoints_list = KeypointsVector()
-        keypoints_list.append(kps_i0)
-        keypoints_list.append(kps_i1)
-        keypoints_list.append(kps_i2)
+        keypoints = KeypointsVector()
+        keypoints.append(kps_i0)
+        keypoints.append(kps_i1)
+        keypoints.append(kps_i2)
 
         # For each image pair (i1,i2), we provide a (K,2) matrix
         # of corresponding image indices (k1,k2).
-        matches_dict = MatchIndicesMap()
-        matches_dict[IndexPair(0, 1)] = np.array([[0, 0], [1, 1]])
-        matches_dict[IndexPair(1, 2)] = np.array([[2, 0], [1, 1]])
+        matches = MatchIndicesMap()
+        matches[IndexPair(0, 1)] = np.array([[0, 0], [1, 1]])
+        matches[IndexPair(1, 2)] = np.array([[2, 0], [1, 1]])
 
         tracks = gtsam.gtsfm.tracksFromPairwiseMatches(
-            matches_dict,
-            keypoints_list,
+            matches,
+            keypoints,
             verbose=False,
         )
         assert len(tracks) == 3
@@ -93,5 +119,71 @@ class TestSfmTrack2d(GtsamTestCase):
         assert track.numberMeasurements() == 1
 
 
+def get_dummy_keypoints_list() -> KeypointsVector:
+    """Generate a list of dummy keypoints for testing."""
+    img1_kp_coords = np.array([[1, 1], [2, 2], [3, 3.]])
+    img2_kp_coords = np.array(
+        [
+            [1, 1.],
+            [2, 2],
+            [3, 3],
+            [4, 4],
+            [5, 5],
+            [6, 6],
+            [7, 7],
+            [8, 8],
+        ]
+    )
+    img3_kp_coords = np.array(
+        [
+            [1, 1.],
+            [2, 2],
+            [3, 3],
+            [4, 4],
+            [5, 5],
+            [6, 6],
+            [7, 7],
+            [8, 8],
+            [9, 9],
+            [10, 10],
+        ]
+    )
+    img4_kp_coords = np.array(
+        [
+            [1, 1.],
+            [2, 2],
+            [3, 3],
+            [4, 4],
+            [5, 5],
+        ]
+    )
+    keypoints = KeypointsVector()
+    keypoints.append(Keypoints(coordinates=img1_kp_coords))
+    keypoints.append(Keypoints(coordinates=img2_kp_coords))
+    keypoints.append(Keypoints(coordinates=img3_kp_coords))
+    keypoints.append(Keypoints(coordinates=img4_kp_coords))
+    return keypoints
+
+
+def get_nontransitive_matches() -> Dict[Tuple[int, int], np.ndarray]:
+    """Set up correspondences for each (i1,i2) pair that violates transitivity.
+
+    (i=0, k=0)             (i=0, k=1)
+         |    \\               |
+         |     \\              |
+    (i=1, k=2)--(i=2,k=3)--(i=3, k=4)
+
+    Transitivity is violated due to the match between frames 0 and 3.
+    """
+    nontransitive_matches = {
+        (0, 1): np.array([[0, 2]]),
+        (1, 2): np.array([[2, 3]]),
+        (0, 2): np.array([[0, 3]]),
+        (0, 3): np.array([[1, 4]]),
+        (2, 3): np.array([[3, 4]]),
+    }
+    return nontransitive_matches
+
+
 if __name__ == "__main__":
     unittest.main()