New experimental camera tracking pipeline

bin/meshroom_batch

@@ -21,7 +21,7 @@ parser.add_argument('-I', '--inputRecursive', metavar='FOLDERS/IMAGES', type=str
                     help='Input folders containing all images recursively.')
 
 parser.add_argument('-p', '--pipeline', metavar='photogrammetry/panoramaHdr/panoramaFisheyeHdr/MG_FILE', type=str, default='photogrammetry',
-                    help='"photogrammetry" pipeline, "panoramaHdr" pipeline, "panoramaFisheyeHdr" pipeline or a Meshroom file containing a custom pipeline to run on input images. '
+                    help='"photogrammetry", "panoramaHdr", "panoramaFisheyeHdr", "cameraTracking" pipeline or a Meshroom file containing a custom pipeline to run on input images. '
                          'Requirements: the graph must contain one CameraInit node, '
                          'and one Publish node if --output is set.')
 
@@ -119,6 +119,9 @@ with multiview.GraphModification(graph):
     elif args.pipeline.lower() == "panoramafisheyehdr":
         # default panorama Fisheye Hdr pipeline
         multiview.panoramaFisheyeHdr(inputViewpoints=views, inputIntrinsics=intrinsics, output=args.output, graph=graph)
+    elif args.pipeline.lower() == "cameratracking":
+        # default panorama Fisheye Hdr pipeline
+        multiview.cameraTracking(inputViewpoints=views, inputIntrinsics=intrinsics, output=args.output, graph=graph)
     else:
         # custom pipeline
         graph.load(args.pipeline)

meshroom/multiview.py

@@ -182,9 +182,11 @@ def panoramaFisheyeHdr(inputImages=None, inputViewpoints=None, inputIntrinsics=N
         panoramaHdr(inputImages, inputViewpoints, inputIntrinsics, output, graph)
         for panoramaInit in graph.nodesOfType("PanoramaInit"):
             panoramaInit.attribute("useFisheye").value = True
-        # when using fisheye images, the overlap between images can be small
-        # and thus requires many features to get enough correspondances for cameras estimation
         for featureExtraction in graph.nodesOfType("FeatureExtraction"):
+            # when using fisheye images, 'sift' performs better than 'dspsift'
+            featureExtraction.attribute("describerTypes").value = ['sift']
+            # when using fisheye images, the overlap between images can be small
+            # and thus requires many features to get enough correspondances for cameras estimation
             featureExtraction.attribute("describerPreset").value = 'high'
     return graph
 
@@ -468,3 +470,60 @@ def sfmAugmentation(graph, sourceSfm, withMVS=False):
         mvsNodes = mvsPipeline(graph, structureFromMotion)
 
     return sfmNodes, mvsNodes
+
+
+def cameraTrackingPipeline(graph):
+    """
+    Instantiate a camera tracking pipeline inside 'graph'.
+
+    Args:
+        graph (Graph/UIGraph): the graph in which nodes should be instantiated
+
+    Returns:
+        list of Node: the created nodes
+    """
+
+    with GraphModification(graph):
+
+        cameraInit, featureExtraction, imageMatching, featureMatching, structureFromMotion = sfmPipeline(graph)
+
+        imageMatching.attribute("nbMatches").value = 5  # voctree nb matches
+        imageMatching.attribute("nbNeighbors").value = 10
+
+        structureFromMotion.attribute("minNumberOfMatches").value = 0
+        structureFromMotion.attribute("minInputTrackLength").value = 5
+        structureFromMotion.attribute("minNumberOfObservationsForTriangulation").value = 3
+        structureFromMotion.attribute("minAngleForTriangulation").value = 1.0
+        structureFromMotion.attribute("minAngleForLandmark").value = 0.5
+
+        exportAnimatedCamera = graph.addNewNode('ExportAnimatedCamera', input=structureFromMotion.output)
+
+    # store current pipeline version in graph header
+    graph.header.update({'pipelineVersion': __version__})
+
+    return [
+        cameraInit,
+        featureExtraction,
+        imageMatching,
+        featureMatching,
+        structureFromMotion,
+        exportAnimatedCamera,
+        ]
+
+
+def cameraTracking(inputImages=list(), inputViewpoints=list(), inputIntrinsics=list(), output='', graph=None):
+    if not graph:
+        graph = Graph('Camera Tracking')
+    with GraphModification(graph):
+        trackingNodes = cameraTrackingPipeline(graph)
+        cameraInit = trackingNodes[0]
+        cameraInit.viewpoints.extend([{'path': image} for image in inputImages])
+        cameraInit.viewpoints.extend(inputViewpoints)
+        cameraInit.intrinsics.extend(inputIntrinsics)
+
+        if output:
+            exportNode = trackingNodes[-1]
+            graph.addNewNode('Publish', output=output, inputFiles=[exportNode.output])
+
+    return graph
+

meshroom/nodes/aliceVision/CameraLocalization.py

@@ -42,7 +42,7 @@ class CameraLocalization(desc.CommandLineNode):
             name='matchDescTypes',
             label='Match Desc Types',
             description='''Describer types to use for the matching.''',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
             exclusive=False,
             uid=[0],

meshroom/nodes/aliceVision/CameraRigCalibration.py

@@ -49,7 +49,7 @@ class CameraRigCalibration(desc.CommandLineNode):
             name='matchDescTypes',
             label='Match Describer Types',
             description='''The describer types to use for the matching''',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
             exclusive=False,
             uid=[0],

meshroom/nodes/aliceVision/CameraRigLocalization.py

@@ -49,8 +49,8 @@ class CameraRigLocalization(desc.CommandLineNode):
             name='matchDescTypes',
             label='Match Describer Types',
             description='''The describer types to use for the matching''',
-            value=['sift'],
-            values=['sift', 'sift_float', 'sift_upright', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
+            value=['dspsift'],
+            values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
             exclusive=False,
             uid=[0],
             joinChar=',',

meshroom/nodes/aliceVision/ConvertSfMFormat.py

@@ -35,7 +35,7 @@ class ConvertSfMFormat(desc.CommandLineNode):
             name='describerTypes',
             label='Describer Types',
             description='Describer types to keep.',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv', 'unknown'],
             exclusive=False,
             uid=[0],

meshroom/nodes/aliceVision/ExportMatches.py

@@ -21,7 +21,7 @@ class ExportMatches(desc.CommandLineNode):
             name='describerTypes',
             label='Describer Types',
             description='Describer types used to describe an image.',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
             exclusive=False,
             uid=[0],

meshroom/nodes/aliceVision/FeatureExtraction.py

@@ -42,7 +42,7 @@ class FeatureExtraction(desc.CommandLineNode):
             name='describerTypes',
             label='Describer Types',
             description='Describer types used to describe an image.',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
             exclusive=False,
             uid=[0],

meshroom/nodes/aliceVision/FeatureMatching.py

@@ -63,7 +63,7 @@ class FeatureMatching(desc.CommandLineNode):
             name='describerTypes',
             label='Describer Types',
             description='Describer types used to describe an image.',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
             exclusive=False,
             uid=[0],

meshroom/nodes/aliceVision/GlobalSfM.py

@@ -52,7 +52,7 @@ class GlobalSfM(desc.CommandLineNode):
             name='describerTypes',
             label='Describer Types',
             description='Describer types used to describe an image.',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4',
                     'sift_ocv', 'akaze_ocv'],
             exclusive=False,

meshroom/nodes/aliceVision/ImageMatching.py

@@ -65,7 +65,7 @@ class ImageMatching(desc.CommandLineNode):
             ' * Exhaustive: Export all image pairs.\n'
             ' * Frustum: If images have known poses, computes the intersection between cameras frustums to create the list of image pairs.\n'
             ' * FrustumOrVocabularyTree: If images have known poses, use frustum intersection else use VocabularyTree.\n',
-            value='VocabularyTree',
+            value='SequentialAndVocabularyTree',
             values=['VocabularyTree', 'Sequential', 'SequentialAndVocabularyTree', 'Exhaustive', 'Frustum', 'FrustumOrVocabularyTree'],
             exclusive=True,
             uid=[0],
@@ -111,7 +111,7 @@ class ImageMatching(desc.CommandLineNode):
             name='nbMatches',
             label='Voc Tree: Nb Matches',
             description='The number of matches to retrieve for each image (If 0 it will retrieve all the matches).',
-            value=50,
+            value=40,
             range=(0, 1000, 1),
             uid=[0],
             advanced=True,
@@ -121,7 +121,7 @@ class ImageMatching(desc.CommandLineNode):
             name='nbNeighbors',
             label='Sequential: Nb Neighbors',
             description='The number of neighbors to retrieve for each image (If 0 it will retrieve all the neighbors).',
-            value=50,
+            value=5,
             range=(0, 1000, 1),
             uid=[0],
             advanced=True,

meshroom/nodes/aliceVision/KeyframeSelection.py

@@ -109,14 +109,14 @@ class KeyframeSelection(desc.CommandLineNode):
             name='useSparseDistanceSelection',
             label='Use Sparse Distance Selection',
             description='Use sparseDistance selection in order to avoid similar keyframes.',
-            value=True,
+            value=False,
             uid=[0],
         ),
         desc.BoolParam(
             name='useSharpnessSelection',
             label='Use Sharpness Selection',
             description='Use frame sharpness score for keyframe selection.',
-            value=True,
+            value=False,
             uid=[0],
         ),
         desc.FloatParam(
@@ -148,15 +148,15 @@ class KeyframeSelection(desc.CommandLineNode):
             name='minFrameStep',
             label='Min Frame Step',
             description='''minimum number of frames between two keyframes''',
-            value=12,
+            value=1,
             range=(1, 100, 1),
             uid=[0],
         ),
         desc.IntParam(
             name='maxFrameStep',
             label='Max Frame Step',
             description='''maximum number of frames after which a keyframe can be taken''',
-            value=36,
+            value=2,
             range=(2, 1000, 1),
             uid=[0],
         ),

meshroom/nodes/aliceVision/StructureFromMotion.py

@@ -97,7 +97,7 @@ class StructureFromMotion(desc.CommandLineNode):
             name='describerTypes',
             label='Describer Types',
             description='Describer types used to describe an image.',
-            value=['sift'],
+            value=['dspsift'],
             values=['sift', 'sift_float', 'sift_upright', 'dspsift', 'akaze', 'akaze_liop', 'akaze_mldb', 'cctag3', 'cctag4', 'sift_ocv', 'akaze_ocv'],
             exclusive=False,
             uid=[0],
@@ -119,7 +119,7 @@ class StructureFromMotion(desc.CommandLineNode):
             description='Observation contraint mode used in the optimization:\n'
                         ' * Basic: Use standard reprojection error in pixel coordinates\n'
                         ' * Scale: Use reprojection error in pixel coordinates but relative to the feature scale',
-            value='Basic',
+            value='Scale',
             values=['Basic', 'Scale'],
             exclusive=True,
             uid=[0],

meshroom/ui/qml/main.qml

@@ -420,6 +420,10 @@ ApplicationWindow {
                     text: "Panorama Fisheye HDR"
                     onTriggered: ensureSaved(function() { _reconstruction.new("panoramafisheyehdr") })
                 }
+                Action {
+                    text: "Camera Tracking (experimental)"
+                    onTriggered: ensureSaved(function() { _reconstruction.new("cameratracking") })
+                }
             }
             Action {
                 id: openActionItem

meshroom/ui/reconstruction.py

@@ -490,6 +490,9 @@ def new(self, pipeline=None):
         elif p.lower() == "panoramafisheyehdr":
             # default panorama fisheye hdr pipeline
             self.setGraph(multiview.panoramaFisheyeHdr())
+        elif p.lower() == "cameratracking":
+            # default camera tracking pipeline
+            self.setGraph(multiview.cameraTracking())
         else:
             # use the user-provided default photogrammetry project file
             self.load(p, setupProjectFile=False)