KITTI 00 sequence & rtabmap.

[Mikor]

Hi Mathieu,

I have some questions about your results that I've read at this paper. I want to replicate what you've done at the sequence 00 using only lidar. I've found out that at this post, as you say, you're passing the arguments that were used for the tests that were made in the paper. So, I launched rtabmap passing these arguments.

roslaunch rtabmap_ros rtabmap.launch    \
   use_sim_time:=true \
   depth:=false \
   subscribe_scan_cloud:=true \
   frame_id:=base_link \
   scan_cloud_topic:=/kitti/velo/pointcloud  \
   scan_cloud_max_points:=131072  \
   icp_odometry:=true \
   approx_sync:=false \
   args:="-d  \
      --RGBD/CreateOccupancyGrid false \
      --Rtabmap/DetectionRate 2 \
      --Odom/ScanKeyFrameThr 0.8 \
      --OdomF2M/ScanMaxSize 10000  \
      --OdomF2M/ScanSubtractRadius 0.5   \
      --Icp/PM true \
      --Icp/VoxelSize 0.5   \
      --Icp/MaxTranslation 2   \
      --Icp/MaxCorrespondenceDistance 1.5 \
      --Icp/PMOutlierRatio 0.7 \
      --Icp/Iterations 10 \
      --Icp/PointToPlane false \
      --Icp/PMMatcherKnn 3 \
      --Icp/PMMatcherEpsilon 1 \
      --Icp/Epsilon 0.0001 \
      --Icp/PointToPlaneK 0 \
      --Icp/PointToPlaneRadius 0 \
      --Icp/CorrespondenceRatio 0.01"

Also, as I've read from the odometry developement kit the sequence 00 can be downloaded from here according to the following table.

Nr. Sequence name Start End

00: 2011_10_03_drive_0027 000000 004540
01: 2011_10_03_drive_0042 000000 001100
02: 2011_10_03_drive_0034 000000 004660
03: 2011_09_26_drive_0067 000000 000800
04: 2011_09_30_drive_0016 000000 000270
05: 2011_09_30_drive_0018 000000 002760
06: 2011_09_30_drive_0020 000000 001100
07: 2011_09_30_drive_0027 000000 001100
08: 2011_09_30_drive_0028 001100 005170
09: 2011_09_30_drive_0033 000000 001590
10: 2011_09_30_drive_0034 000000 001200

Assuming that I am right at all the previously said, my graph looks like this which is completely different from the results that are shown in the paper, but I don't know what am I doing wrong...



Also, when I use the imu_topic it says that it can not perform the registration with a null guess. So, I do not know what to do to improve my solution. And when I am rerunning the rtabmap I get different results. So Mathieu, do you have any idea for what I've been missing ?

Sincerely,
Anthony.

[matlabbe]

Hi Anthony,

For the paper, the dataset from this page has been used. The tool used to evaluate that dataset is called rtabmap-kitti_dataset, used in that script.

I just tried the command you used with that bag:


This is without loop closure detection. The results in the paper are with loop closure detection. What is your rtabmap config?

$ rtabmap --version

cheers,
Mathieu

[Mikor]

Hey Mathieu,

antonis@antonis-VirtualBox:~$ rtabmap --version
RTAB-Map:               0.20.9
PCL:                     1.8.1
With VTK:                6.3.0
OpenCV:                  3.2.0
With OpenCV xfeatures2d: false
With OpenCV nonfree:     false
With ORB OcTree:          true
With SuperPoint Torch:   false
With Python3:            false
With FastCV:             false
With Madgwick:            true
With TORO:                true
With g2o:                 true
With GTSAM:               true
With Vertigo:             true
With CVSBA:              false
With Ceres:              false
With OpenNI2:             true
With Freenect:            true
With Freenect2:          false
With K4W2:               false
With K4A:                false
With DC1394:              true
With FlyCapture2:        false
With ZED:                false
With RealSense:          false
With RealSense SLAM:     false
With RealSense2:         false
With MYNT EYE S:         false
With libpointmatcher:     true
With octomap:             true
With cpu-tsdf:           false
With open chisel:        false
With Alice Vision:       false
With LOAM:               false
With FOVIS:              false
With Viso2:              false
With DVO:                false
With ORB_SLAM2:          false
With OKVIS:              false
With MSCKF_VIO:          false
With VINS-Fusion:        false

This is my configuration.

[matlabbe]

It looks okay, and the created database:

rtabmap-info ~/.ros/rtabmap.db

[Mikor]

This is my output:

$ rtabmap-info ~/.ros/rtabmap.db
Parameters (Yellow=modified, Red=old parameter not used anymore):
BRIEF/Bytes=                       32
BRISK/Octaves=                     3
BRISK/PatternScale=                1
BRISK/Thresh=                      30
Bayes/FullPredictionUpdate=        false
Bayes/PredictionLC=                0.1 0.36 0.30 0.16 0.062 0.0151 0.00255 0.000324 2.5e-05 1.3e-06 4.8e-08 1.2e-09 1.9e-11 2.2e-13 1.7e-15 8.5e-18 2.9e-20 6.9e-23
Bayes/VirtualPlacePriorThr=        0.9
DbSqlite3/CacheSize=               10000
DbSqlite3/InMemory=                false
DbSqlite3/JournalMode=             3
DbSqlite3/Synchronous=             0
DbSqlite3/TempStore=               2
FAST/CV=                           0
FAST/Gpu=                          false
FAST/GpuKeypointsRatio=            0.05
FAST/GridCols=                     0
FAST/GridRows=                     0
FAST/MaxThreshold=                 200
FAST/MinThreshold=                 7
FAST/NonmaxSuppression=            true
FAST/Threshold=                    20
FREAK/NOctaves=                    4
FREAK/OrientationNormalized=       true
FREAK/PatternScale=                22
FREAK/ScaleNormalized=             true
GFTT/BlockSize=                    3
GFTT/K=                            0.04
GFTT/MinDistance=                  7
GFTT/QualityLevel=                 0.001
GFTT/UseHarrisDetector=            false
GMS/ThresholdFactor=               6.0
GMS/WithRotation=                  false
GMS/WithScale=                     false
GTSAM/Optimizer=                   1
Grid/3D=                           true
Grid/CellSize=                     0.05
Grid/ClusterRadius=                0.1
Grid/DepthDecimation=              4
Grid/DepthRoiRatios=               0.0 0.0 0.0 0.0
Grid/FlatObstacleDetected=         true
Grid/FootprintHeight=              0.0
Grid/FootprintLength=              0.0
Grid/FootprintWidth=               0.0
Grid/FromDepth=                    false  (default=true)
Grid/GroundIsObstacle=             false
Grid/MapFrameProjection=           false
Grid/MaxGroundAngle=               45
Grid/MaxGroundHeight=              0.0
Grid/MaxObstacleHeight=            0.0
Grid/MinClusterSize=               10
Grid/MinGroundHeight=              0.0
Grid/NoiseFilteringMinNeighbors=   5
Grid/NoiseFilteringRadius=         0.0
Grid/NormalK=                      20
Grid/NormalsSegmentation=          true
Grid/PreVoxelFiltering=            true
Grid/RangeMax=                     0  (default=5.0)
Grid/RangeMin=                     0.0
Grid/RayTracing=                   false
Grid/Scan2dUnknownSpaceFilled=     false
Grid/ScanDecimation=               1
GridGlobal/AltitudeDelta=          0
GridGlobal/Eroded=                 false
GridGlobal/FootprintRadius=        0.0
GridGlobal/FullUpdate=             true
GridGlobal/MaxNodes=               0
GridGlobal/MinSize=                0.0
GridGlobal/OccupancyThr=           0.5
GridGlobal/ProbClampingMax=        0.971
GridGlobal/ProbClampingMin=        0.1192
GridGlobal/ProbHit=                0.7
GridGlobal/ProbMiss=               0.4
GridGlobal/UpdateError=            0.01
Icp/CorrespondenceRatio=           0.01  (default=0.1)
Icp/DownsamplingStep=              1
Icp/Epsilon=                       0.0001  (default=0)
Icp/Iterations=                    10  (default=30)
Icp/MaxCorrespondenceDistance=     1.5  (default=0.1)
Icp/MaxRotation=                   0.78
Icp/MaxTranslation=                2  (default=0.2)
Icp/PM=                            true
Icp/PMForce4DoF=                   false
Icp/PMMatcherEpsilon=              1  (default=0.0)
Icp/PMMatcherIntensity=            false
Icp/PMMatcherKnn=                  3  (default=1)
Icp/PMOutlierRatio=                0.7  (default=0.85)
Icp/PointToPlane=                  false  (default=true)
Icp/PointToPlaneGroundNormalsUp=   0.0
Icp/PointToPlaneK=                 0  (default=5)
Icp/PointToPlaneLowComplexityStrategy=1
Icp/PointToPlaneMinComplexity=     0.02
Icp/PointToPlaneRadius=            0  (default=1.0)
Icp/RangeMax=                      0
Icp/RangeMin=                      0
Icp/VoxelSize=                     0.5  (default=0.05)
ImuFilter/ComplementaryBiasAlpha=  0.01
ImuFilter/ComplementaryDoAdpativeGain=true
ImuFilter/ComplementaryDoBiasEstimation=true
ImuFilter/ComplementaryGainAcc=    0.01
ImuFilter/MadgwickGain=            0.1
ImuFilter/MadgwickZeta=            0.0
KAZE/Diffusivity=                  1
KAZE/Extended=                     false
KAZE/NOctaveLayers=                4
KAZE/NOctaves=                     4
KAZE/Threshold=                    0.001
KAZE/Upright=                      false
Kp/BadSignRatio=                   0.5
Kp/ByteToFloat=                    false
Kp/DetectorStrategy=               8
Kp/FlannRebalancingFactor=         2.0
Kp/GridCols=                       1
Kp/GridRows=                       1
Kp/IncrementalDictionary=          true
Kp/IncrementalFlann=               true
Kp/MaxDepth=                       0
Kp/MaxFeatures=                    500
Kp/MinDepth=                       0
Kp/NNStrategy=                     1
Kp/NewWordsComparedTogether=       true
Kp/NndrRatio=                      0.8
Kp/Parallelized=                   true
Kp/RoiRatios=                      0.0 0.0 0.0 0.0
Kp/SubPixEps=                      0.02
Kp/SubPixIterations=               0
Kp/SubPixWinSize=                  3
Kp/TfIdfLikelihoodUsed=            true
Marker/CornerRefinementMethod=     0
Marker/Dictionary=                 0
Marker/Length=                     0
Marker/MaxDepthError=              0.01
Marker/MaxRange=                   0.0
Marker/MinRange=                   0.0
Marker/VarianceAngular=            0.01
Marker/VarianceLinear=             0.001
Mem/BadSignaturesIgnored=          false
Mem/BinDataKept=                   true
Mem/CompressionParallelized=       true
Mem/CovOffDiagIgnored=             true
Mem/DepthAsMask=                   true
Mem/GenerateIds=                   true
Mem/ImageCompressionFormat=        .jpg
Mem/ImageKept=                     false
Mem/ImagePostDecimation=           1
Mem/ImagePreDecimation=            1
Mem/IncrementalMemory=             true
Mem/InitWMWithAllNodes=            false
Mem/IntermediateNodeDataKept=      false
Mem/LaserScanDownsampleStepSize=   1
Mem/LaserScanNormalK=              0
Mem/LaserScanNormalRadius=         0.0
Mem/LaserScanVoxelSize=            0.0
Mem/LocalizationDataSaved=         false
Mem/MapLabelsAdded=                true
Mem/NotLinkedNodesKept=            true
Mem/RawDescriptorsKept=            true
Mem/RecentWmRatio=                 0.2
Mem/ReduceGraph=                   false
Mem/RehearsalIdUpdatedToNewOne=    false
Mem/RehearsalSimilarity=           0.6
Mem/RehearsalWeightIgnoredWhileMoving=false
Mem/STMSize=                       10
Mem/SaveDepth16Format=             false
Mem/StereoFromMotion=              false
Mem/TransferSortingByWeightId=     false
Mem/UseOdomFeatures=               true
Mem/UseOdomGravity=                false
ORB/EdgeThreshold=                 19
ORB/FirstLevel=                    0
ORB/Gpu=                           false
ORB/NLevels=                       3
ORB/PatchSize=                     31
ORB/ScaleFactor=                   2
ORB/ScoreType=                     0
ORB/WTA_K=                         2
Odom/AlignWithGround=              false
Odom/FillInfoData=                 true
Odom/FilteringStrategy=            0
Odom/GuessMotion=                  true
Odom/GuessSmoothingDelay=          0
Odom/Holonomic=                    true
Odom/ImageBufferSize=              1
Odom/ImageDecimation=              1
Odom/KalmanMeasurementNoise=       0.01
Odom/KalmanProcessNoise=           0.001
Odom/KeyFrameThr=                  0.3
Odom/ParticleLambdaR=              100
Odom/ParticleLambdaT=              100
Odom/ParticleNoiseR=               0.002
Odom/ParticleNoiseT=               0.002
Odom/ParticleSize=                 400
Odom/ResetCountdown=               0
Odom/ScanKeyFrameThr=              0.8  (default=0.9)
Odom/Strategy=                     0
Odom/VisKeyFrameThr=               150
OdomF2M/BundleAdjustment=          1
OdomF2M/BundleAdjustmentMaxFrames= 10
OdomF2M/MaxNewFeatures=            0
OdomF2M/MaxSize=                   2000
OdomF2M/ScanMaxSize=               10000  (default=2000)
OdomF2M/ScanRange=                 0
OdomF2M/ScanSubtractAngle=         45
OdomF2M/ScanSubtractRadius=        0.5  (default=0.05)
OdomF2M/ValidDepthRatio=           0.75
OdomFovis/BucketHeight=            80
OdomFovis/BucketWidth=             80
OdomFovis/CliqueInlierThreshold=   0.1
OdomFovis/FastThreshold=           20
OdomFovis/FastThresholdAdaptiveGain=0.005
OdomFovis/FeatureSearchWindow=     25
OdomFovis/FeatureWindowSize=       9
OdomFovis/InlierMaxReprojectionError=1.5
OdomFovis/MaxKeypointsPerBucket=   25
OdomFovis/MaxMeanReprojectionError=10.0
OdomFovis/MaxPyramidLevel=         3
OdomFovis/MinFeaturesForEstimate=  20
OdomFovis/MinPyramidLevel=         0
OdomFovis/StereoMaxDisparity=      128
OdomFovis/StereoMaxDistEpipolarLine=1.5
OdomFovis/StereoMaxRefinementDisplacement=1.0
OdomFovis/StereoRequireMutualMatch=true
OdomFovis/TargetPixelsPerFeature=  250
OdomFovis/UpdateTargetFeaturesWithRefined=false
OdomFovis/UseAdaptiveThreshold=    true
OdomFovis/UseBucketing=            true
OdomFovis/UseHomographyInitialization=true
OdomFovis/UseImageNormalization=   false
OdomFovis/UseSubpixelRefinement=   true
OdomLOAM/AngVar=                   0.01
OdomLOAM/LinVar=                   0.01
OdomLOAM/LocalMapping=             true
OdomLOAM/ScanPeriod=               0.1
OdomLOAM/Sensor=                   2
OdomMSCKF/FastThreshold=           10
OdomMSCKF/GridCol=                 5
OdomMSCKF/GridMaxFeatureNum=       4
OdomMSCKF/GridMinFeatureNum=       3
OdomMSCKF/GridRow=                 4
OdomMSCKF/InitCovAccBias=          0.01
OdomMSCKF/InitCovExRot=            0.00030462
OdomMSCKF/InitCovExTrans=          0.000025
OdomMSCKF/InitCovGyroBias=         0.01
OdomMSCKF/InitCovVel=              0.25
OdomMSCKF/MaxCamStateSize=         20
OdomMSCKF/MaxIteration=            30
OdomMSCKF/NoiseAcc=                0.05
OdomMSCKF/NoiseAccBias=            0.01
OdomMSCKF/NoiseFeature=            0.035
OdomMSCKF/NoiseGyro=               0.005
OdomMSCKF/NoiseGyroBias=           0.001
OdomMSCKF/OptTranslationThreshold= 0
OdomMSCKF/PatchSize=               15
OdomMSCKF/PositionStdThreshold=    8.0
OdomMSCKF/PyramidLevels=           3
OdomMSCKF/RansacThreshold=         3
OdomMSCKF/RotationThreshold=       0.2618
OdomMSCKF/StereoThreshold=         5
OdomMSCKF/TrackPrecision=          0.01
OdomMSCKF/TrackingRateThreshold=   0.5
OdomMSCKF/TranslationThreshold=    0.4
OdomMono/InitMinFlow=              100
OdomMono/InitMinTranslation=       0.1
OdomMono/MaxVariance=              0.01
OdomMono/MinTranslation=           0.02
OdomORBSLAM2/Bf=                   0.076
OdomORBSLAM2/Fps=                  0.0
OdomORBSLAM2/MapSize=              3000
OdomORBSLAM2/MaxFeatures=          1000
OdomORBSLAM2/ThDepth=              40.0
OdomViso2/BucketHeight=            50
OdomViso2/BucketMaxFeatures=       2
OdomViso2/BucketWidth=             50
OdomViso2/InlierThreshold=         2.0
OdomViso2/MatchBinsize=            50
OdomViso2/MatchDispTolerance=      2
OdomViso2/MatchHalfResolution=     true
OdomViso2/MatchMultiStage=         true
OdomViso2/MatchNmsN=               3
OdomViso2/MatchNmsTau=             50
OdomViso2/MatchOutlierDispTolerance=5
OdomViso2/MatchOutlierFlowTolerance=5
OdomViso2/MatchRadius=             200
OdomViso2/MatchRefinement=         1
OdomViso2/RansacIters=             200
OdomViso2/Reweighting=             true
Optimizer/Epsilon=                 0.00001
Optimizer/GravitySigma=            0.3
Optimizer/Iterations=              20
Optimizer/LandmarksIgnored=        false
Optimizer/PriorsIgnored=           true
Optimizer/Robust=                  false
Optimizer/Strategy=                2
Optimizer/VarianceIgnored=         false
PyDetector/Cuda=                   true
PyMatcher/Cuda=                    true
PyMatcher/Iterations=              20
PyMatcher/Model=                   indoor
PyMatcher/Threshold=               0.2
RGBD/AngularSpeedUpdate=           0.0
RGBD/AngularUpdate=                0.1
RGBD/CreateOccupancyGrid=          false
RGBD/Enabled=                      true
RGBD/GoalReachedRadius=            0.5
RGBD/GoalsSavedInUserData=         false
RGBD/LinearSpeedUpdate=            0.0
RGBD/LinearUpdate=                 0.1
RGBD/LocalBundleOnLoopClosure=     false
RGBD/LocalImmunizationRatio=       0.25
RGBD/LocalRadius=                  10
RGBD/LoopClosureReextractFeatures= false
RGBD/LoopCovLimited=               false
RGBD/MarkerDetection=              false
RGBD/MaxLocalRetrieved=            2
RGBD/MaxLoopClosureDistance=       0.0
RGBD/MaxOdomCacheSize=             0
RGBD/NeighborLinkRefining=         false
RGBD/NewMapOdomChangeDistance=     0
RGBD/OptimizeFromGraphEnd=         false
RGBD/OptimizeMaxError=             3.0
RGBD/PlanAngularVelocity=          0
RGBD/PlanLinearVelocity=           0
RGBD/PlanStuckIterations=          0
RGBD/ProximityAngle=               45
RGBD/ProximityBySpace=             true
RGBD/ProximityByTime=              false
RGBD/ProximityMaxGraphDepth=       50
RGBD/ProximityMaxPaths=            3
RGBD/ProximityOdomGuess=           false
RGBD/ProximityPathFilteringRadius= 1
RGBD/ProximityPathMaxNeighbors=    0
RGBD/ProximityPathRawPosesUsed=    true
RGBD/SavedLocalizationIgnored=     false
RGBD/ScanMatchingIdsSavedInLinks=  true
Reg/Force3DoF=                     false
Reg/RepeatOnce=                    true
Reg/Strategy=                      0
Rtabmap/ComputeRMSE=               true
Rtabmap/CreateIntermediateNodes=   false
Rtabmap/DetectionRate=             2  (default=1)
Rtabmap/ImageBufferSize=           1
Rtabmap/ImagesAlreadyRectified=    true
Rtabmap/LoopGPS=                   true
Rtabmap/LoopRatio=                 0
Rtabmap/LoopThr=                   0.11
Rtabmap/MaxRetrieved=              2
Rtabmap/MemoryThr=                 0
Rtabmap/PublishLastSignature=      true
Rtabmap/PublishLikelihood=         true
Rtabmap/PublishPdf=                true
Rtabmap/PublishRAMUsage=           false
Rtabmap/PublishStats=              true
Rtabmap/RectifyOnlyFeatures=       false
Rtabmap/SaveWMState=               false
Rtabmap/StartNewMapOnGoodSignature=false
Rtabmap/StartNewMapOnLoopClosure=  false
Rtabmap/StatisticLogged=           false
Rtabmap/StatisticLoggedHeaders=    true
Rtabmap/StatisticLogsBufferedInRAM=true
Rtabmap/TimeThr=                   0
SIFT/ContrastThreshold=            0.04
SIFT/EdgeThreshold=                10
SIFT/NFeatures=                    0
SIFT/NOctaveLayers=                3
SIFT/RootSIFT=                     false
SIFT/Sigma=                        1.6
SURF/Extended=                     false
SURF/GpuKeypointsRatio=            0.01
SURF/GpuVersion=                   false
SURF/HessianThreshold=             500
SURF/OctaveLayers=                 2
SURF/Octaves=                      4
SURF/Upright=                      false
Stereo/DenseStrategy=              0
Stereo/Eps=                        0.01
Stereo/Iterations=                 30
Stereo/MaxDisparity=               128.0
Stereo/MaxLevel=                   5
Stereo/MinDisparity=               0.5
Stereo/OpticalFlow=                true
Stereo/SSD=                        true
Stereo/WinHeight=                  3
Stereo/WinWidth=                   15
StereoBM/BlockSize=                15
StereoBM/Disp12MaxDiff=            -1
StereoBM/MinDisparity=             0
StereoBM/NumDisparities=           128
StereoBM/PreFilterCap=             31
StereoBM/PreFilterSize=            9
StereoBM/SpeckleRange=             4
StereoBM/SpeckleWindowSize=        100
StereoBM/TextureThreshold=         10
StereoBM/UniquenessRatio=          15
StereoSGBM/BlockSize=              15
StereoSGBM/Disp12MaxDiff=          1
StereoSGBM/MinDisparity=           0
StereoSGBM/Mode=                   2
StereoSGBM/NumDisparities=         128
StereoSGBM/P1=                     2
StereoSGBM/P2=                     5
StereoSGBM/PreFilterCap=           31
StereoSGBM/SpeckleRange=           4
StereoSGBM/SpeckleWindowSize=      100
StereoSGBM/UniquenessRatio=        20
SuperPoint/Cuda=                   true
SuperPoint/NMS=                    true
SuperPoint/NMSRadius=              4
SuperPoint/Threshold=              0.010
VhEp/Enabled=                      false
VhEp/MatchCountMin=                8
VhEp/RansacParam1=                 3
VhEp/RansacParam2=                 0.99
Vis/BundleAdjustment=              1
Vis/CorFlowEps=                    0.01
Vis/CorFlowIterations=             30
Vis/CorFlowMaxLevel=               3
Vis/CorFlowWinSize=                16
Vis/CorGuessMatchToProjection=     false
Vis/CorGuessWinSize=               40
Vis/CorNNDR=                       0.8
Vis/CorNNType=                     1
Vis/CorType=                       0
Vis/DepthAsMask=                   true
Vis/EpipolarGeometryVar=           0.1
Vis/EstimationType=                1
Vis/FeatureType=                   8
Vis/ForwardEstOnly=                true
Vis/GridCols=                      1
Vis/GridRows=                      1
Vis/InlierDistance=                0.1
Vis/Iterations=                    300
Vis/MaxDepth=                      0
Vis/MaxFeatures=                   1000
Vis/MeanInliersDistance=           0.0
Vis/MinDepth=                      0
Vis/MinInliers=                    20
Vis/MinInliersDistribution=        0.0
Vis/PnPFlags=                      0
Vis/PnPRefineIterations=           0
Vis/PnPReprojError=                2
Vis/RefineIterations=              5
Vis/RoiRatios=                     0.0 0.0 0.0 0.0
Vis/SubPixEps=                     0.02
Vis/SubPixIterations=              0
Vis/SubPixWinSize=                 3
g2o/Baseline=                      0.075
g2o/Optimizer=                     0
g2o/PixelVariance=                 1.0
g2o/RobustKernelDelta=             8
g2o/Solver=                        0

Info:

Path:               /home/antonis/.ros/rtabmap.db
Version:            0.20.9
Sessions:           1
Total odometry length:2894.902588 m
Total time:         470.685171s
LTM:                582 nodes and 0 words (dim=0 type=8U)
WM:                 582 nodes and 0 words
Global graph:       582 poses and 605 links
Optimized graph:    0 poses
Maps in graph:      0/1 []
Ground truth:       0 poses
GPS:                0 poses
Links:
  Neighbor:         580
  GlobalClosure:    0
  LocalSpaceClosure:25
  LocalTimeClosure: 0
  UserClosure:      0
  VirtualClosure:   0
  NeighborMerged:   0
  PosePrior:        0
  Landmark:         0
  Gravity:          0

Database size:      551 MB
Nodes size:         92 KB	(0.02%)
Links size:         208 KB	(0.04%)
RGB Images size:    0 Bytes	(0.00%)
Depth Images size:  0 Bytes	(0.00%)
Calibrations size:  0 Bytes	(0.00%)
Grids size:         6 KB	(0.00%)
Scans size:         549 MB	(99.53%)
User data size:     0 Bytes	(0.00%)
Dictionary size:    0 Bytes	(0.00%)
Features size:      0 Bytes	(0.00%)
Statistics size:    835 KB	(0.15%)
Other (indexing, unused):1 MB	(0.26%)

It really makes me wonder what is going wrong, I've downloaded again the data and used the kitti2bag to turn them in bagfiles. Using the

kitti2bag -t <date> -r <sequence number> raw_synced

 command thinking that my data are problematic because rtab works great with the ouster drive under the rain bag and the other that I shared at this post  but still, the odometry goes wild.

[matlabbe]

Hi Mikor,

Here my stats:

Info:

Path:               /home/mathieu/.ros/rtabmap.db
Version:            0.20.9
Sessions:           1
Total odometry length:3710.642822 m
Total time:         470.581516s
LTM:                909 nodes and 0 words (dim=0 type=8U)
WM:                 907 nodes and 0 words
Global graph:       909 poses and 1818 links
Optimized graph:    907 poses
Maps in graph:      1/1 [0(907)]
Ground truth:       0 poses
GPS:                0 poses
Links:
  Neighbor:         1812
  GlobalClosure:    0
  LocalSpaceClosure:6
  LocalTimeClosure: 0
  UserClosure:      0
  VirtualClosure:   0
  NeighborMerged:   0
  PosePrior:        0
  Landmark:         0
  Gravity:          0

Database size:      863 MB
Nodes size:         145 KB	(0.02%)
Links size:         623 KB	(0.07%)
RGB Images size:    0 Bytes	(0.00%)
Depth Images size:  0 Bytes	(0.00%)
Calibrations size:  0 Bytes	(0.00%)
Grids size:         10 KB	(0.00%)
Scans size:         858 MB	(99.49%)
User data size:     0 Bytes	(0.00%)
Dictionary size:    0 Bytes	(0.00%)
Features size:      0 Bytes	(0.00%)
Statistics size:    1 MB	(0.15%)
Other (indexing, unused):2 MB	(0.27%)

I would expect 900-940 nodes in your map (for 2 Hz during 470 sec). Can you show processing time for odometry?

rtabmap-report Odometry/TimeEstimation/ms ~/.ros/rtabmap.db

On my laptop it is able to process all velodyne point clouds (< 100 ms processing time).

[Mikor]

Good morning Mathieu,

I guess it's morning there, here are my results. It averages at 338ms



EDIT: Also I noticed this when I launched rtabmap

[ INFO] [1615558080.547552628]: Odometry: Ignored parameter "Rtabmap/DetectionRate"="2" from arguments

I do not know if this is expected behavior.

And at the end I got these warnings

[ WARN] (2021-03-12 16:19:14.925) MainWindow.cpp:1775::processStats() Processing time (1,380790s) is over detection rate (0,500000s), real-time problem!

[matlabbe]

Hi Mikor,

It is indeed morning here. What are the specifications of your computer? To get similar results than in the paper, it is assumed that the computer can process all the velodyne point clouds (at 10Hz) in real-time.

A workaround would be play slower the bag:

rosbag play -r 0.5 --clock ...

If you are using rtabmapviz to visualize, remove the 3D map rendering to save computation time (Preferences->3D Rendering-> under Map column scroll down to disable scan). On my computer I disable it so that rendering doesn't take much processing power, keeping only odom scan visualization.

EDIT: the warning "Odometry: Ignored parameter "Rtabmap/DetectionRate"="2" from arguments" is fine, that parameter is not used in odometry.

cheers,
Mathieu

[Mikor]

Hey Mathieu,

I tried it with 0.5 rate as you said and it didn't work well enough, now I've set it at 0.1, it takes ages but it looks good so far.

I'm running it on a lenovo legion y530 with an i5-8300H and 8gb of ram, but it's also running on VM using half of the cpu and ram.

I will keep you updated once it's done, thanks once again.

Here it is..!

[matlabbe]

Looks like what I had! So it was a computation power issue.

cheers,
Mathieu

[Mikor]

Hello Mathieu,

I have a KITTI related question so I will post it here, if that's not appropriate I can create a new post. My question is specifically about the evaluation of 07 sequence but I guess it happens to the rest sequences too. Could you help me figure out how to use the evaluation script you posted before, I am not familiar with bash files and I wonder how can the 220 poses, that I get as output, be compared to the 1100 grountruth poses.

Also, is it possible the imu topic to rotate the graph?


The left is with the imu topic.


Thank you in advance Mathieu,

Regards,
Anthony.

[matlabbe]

Hi Anthony,

Do you want to compare trajectory created with the rosbag against the ground truth of the odometry (files not rosbag) leaderboard? For the ground truth, the number of poses shold be exactly the same than the trajectory size. I would also doubt that the poses from the rosbag would be exactly the ground truth poses at exact timestamp. For benchmarking with the kitti's groundtruth, use the folder of files, not the rosbag. Well, you would lose IMU data when not using the rosbag.

You may check if someone has integrated the ground truth in the rosbag. Without the ground truth, you could compare with the GPS at least for qualitative results.

For the bash file, it is using the rtabmap-kitti_dataset tool, here is the description and an example at the bottom (note the Rtabmap/CreateIntermediateNodes that is required to get the same number of poses than the images in the kitti folders, and thus the same number of poses than the ground truth):

$ rtabmap-kitti_dataset 

Usage:
rtabmap-kitti_dataset [options] path
  path               Folder of the sequence (e.g., "~/KITTI/dataset/sequences/07")
                        containing least calib.txt, times.txt, image_0 and image_1 folders.
                        Optional image_2, image_3 and velodyne folders.
  --output           Output directory. By default, results are saved in "path".
  --output_name      Output database name (default "rtabmap").
  --gt "path"        Ground truth path (e.g., ~/KITTI/devkit/cpp/data/odometry/poses/07.txt)
  --quiet            Don't show log messages and iteration updates.
  --color            Use color images for stereo (image_2 and image_3 folders).
  --height           Add car's height to camera local transform (1.67m).
  --disp             Generate full disparity.
  --exposure_comp    Do exposure compensation between left and right images.
  --scan             Include velodyne scan in node's data (use --scan_only to ignore image data).
  --scan_step #      Scan downsample step (default=1).
  --scan_voxel #.#   Scan voxel size (default 0.5 m).
  --scan_k           Scan normal K (default 0).
  --scan_radius      Scan normal radius (default 0).

RTAB-Map options:
   --help                         Show usage.
   --version                      Show version of rtabmap and its dependencies.
   --params                       Show all parameters with their default value and description. 
                                  If a database path is set as last argument, the parameters in the 
                                  database will be shown in INI format.
   --"parameter name" "value"     Overwrite a specific RTAB-Map's parameter :
                                    --SURF/HessianThreshold 150
                                   For parameters in table format, add ',' between values :
                                    --Kp/RoiRatios 0,0,0.1,0
Logger options:
   --nolog              Disable logger
   --logconsole         Set logger console type
   --logfile "path"     Set logger file type
   --logfilea "path"    Set logger file type with appending mode if the file already exists
   --udebug             Set logger level to debug
   --uinfo              Set logger level to info
   --uwarn              Set logger level to warn
   --uerror             Set logger level to error
   --logtime "bool"     Print time when logging
   --logwhere "bool"    Print where when logging
   --logthread "bool"   Print thread id when logging

Example:

   $ rtabmap-kitti_dataset \
       --Rtabmap/PublishRAMUsage true\
       --Rtabmap/DetectionRate 2\
       --Rtabmap/CreateIntermediateNodes true\
       --RGBD/LinearUpdate 0\
       --GFTT/QualityLevel 0.01\
       --GFTT/MinDistance 7\
       --OdomF2M/MaxSize 3000\
       --Mem/STMSize 30\
       --Kp/MaxFeatures 750\
       --Vis/MaxFeatures 1500\
       --gt "~/KITTI/devkit/cpp/data/odometry/poses/07.txt"\
       ~/KITTI/dataset/sequences/07

For the graph rotating with imu, do you use Optimizer/Strategy=2 (GTSAM)? and which rtabmap version are you using? as I added some new constraints to gtsam recently to avoid this kind of rotation. I'll download the rosbag to see if I have the same rotation.

EDIT: I tested the bag, the orientation is coming from the IMU's initial yaw:

[ WARN] (2021-03-16 19:09:13.297) Odometry.cpp:307::process() Updated initial pose from xyz=0.000000,0.000000,0.000000 rpy=0.000000,-0.000000,0.000000 to xyz=0.000000,0.000000,0.000000 rpy=0.022268,0.011411,-0.551108 with IMU orientation

cheers,
Mathieu

[Mikor]

Hi Mathieu,

I try to use the rtabmap-kitti_dataset tool with scan only option enabled.

rtabmap-kitti_dataset \
      --scan_only	\
      --Reg/Strategy 1 \
      --RGBD/Enabled false\
      --RGBD/CreateOccupancyGrid false \
      --Rtabmap/DetectionRate 2 \
      --Odom/ScanKeyFrameThr 0.8 \
      --OdomF2M/ScanMaxSize 10000  \
      --OdomF2M/ScanSubtractRadius 0.5   \
      --Icp/PM true \
      --Icp/VoxelSize 0.5   \
      --Icp/MaxTranslation 2   \
      --Icp/MaxCorrespondenceDistance 1.5 \
      --Icp/PMOutlierRatio 0.7 \
      --Icp/Iterations 10 \
      --Icp/PointToPlane false \
      --Icp/PMMatcherKnn 3 \
      --Icp/PMMatcherEpsilon 1 \
      --Icp/Epsilon 0.0001 \
      --Icp/PointToPlaneK 0 \
      --Icp/PointToPlaneRadius 0 \
      --icp_odometry:=true \
      --Icp/CorrespondenceRatio 0.01 \
      --gt "/home/antonis/07/07.txt" \
      /home/antonis/07

This is the parameters I set but I get this output

Paths:
   Sequence number:  07
   Sequence path:    /home/antonis/07
   Output:           /home/antonis/07
   Output name:      rtabmap
   left images:      /home/antonis/07/image_0
   right images:     /home/antonis/07/image_1
   calib.txt:        /home/antonis/07/calib.txt
   times.txt:        /home/antonis/07/times.txt
   Ground Truth:      /home/antonis/07/07.txt
   Exposure Compensation: false
   Disparity:         false
   Scan:               /home/antonis/07/velodyne
   Scan only:          true
   Scan step:          1
   Scan voxel:         0.500000m
   Scan normal k:      0
   Scan normal radius: 0.000000
Saved calibration "rtabmap_calib" to "/home/antonis/07"
Parameters:
   Icp/CorrespondenceRatio=0.01
   Icp/Epsilon=0.0001
   Icp/Iterations=10
   Icp/MaxCorrespondenceDistance=1.5
   Icp/MaxTranslation=2
   Icp/PM=true
   Icp/PMMatcherEpsilon=1
   Icp/PMMatcherKnn=3
   Icp/PMOutlierRatio=0.7
   Icp/PointToPlane=false
   Icp/PointToPlaneK=0
   Icp/PointToPlaneRadius=0
   Icp/VoxelSize=0.5
   Odom/ScanKeyFrameThr=0.8
   OdomF2M/ScanMaxSize=10000
   OdomF2M/ScanSubtractRadius=0.5
   RGBD/CreateOccupancyGrid=false
   RGBD/Enabled=false
   Reg/Strategy=1
   Rtabmap/DetectionRate=2
   Rtabmap/PublishRAMUsage=true
   Rtabmap/WorkingDirectory=/home/antonis/07
RTAB-Map version: 0.20.9
Processing 1101 images...
[ WARN] (2021-03-17 14:26:03.964) OdometryF2M.cpp:164::OdometryF2M() OdomF2M/BundleAdjustment=1 cannot be used with registration not done only with images (Reg/Strategy=1), disabling bundle adjustment.
Iteration 1/1101: speed=0km/h camera=31ms, odom(quality=0.000000, kfs=1)=0ms, slam=25ms, rmse=0.000000m
Iteration 2/1101: speed=3km/h camera=34ms, odom(quality=0.943217, kfs=1)=154ms, slam=0ms, rmse=0.000000m
Iteration 3/1101: speed=2km/h camera=17ms, odom(quality=0.914549, kfs=1)=122ms, slam=0ms, rmse=0.000000m
Iteration 4/1101: speed=2km/h camera=28ms, odom(quality=0.918854, kfs=1)=132ms, slam=0ms, rmse=0.000000m
Iteration 5/1101: speed=6km/h camera=34ms, odom(quality=0.917901, kfs=1)=129ms, slam=0ms, rmse=0.000000m
Iteration 6/1101: speed=3km/h camera=27ms, odom(quality=0.892217, kfs=1)=125ms, slam=7ms, rmse=0.000000m
Iteration 7/1101: speed=3km/h camera=16ms, odom(quality=0.907208, kfs=1)=115ms, slam=0ms, rmse=0.000000m
Iteration 8/1101: speed=5km/h camera=27ms, odom(quality=0.910995, kfs=1)=100ms, slam=0ms, rmse=0.000000m
Iteration 9/1101: speed=5km/h camera=21ms, odom(quality=0.896019, kfs=1)=100ms, slam=0ms, rmse=0.000000m
Iteration 10/1101: speed=5km/h camera=26ms, odom(quality=0.882009, kfs=1)=97ms, slam=0ms, rmse=0.000000m
Iteration 11/1101: speed=5km/h camera=23ms, odom(quality=0.873261, kfs=1)=128ms, slam=14ms, rmse=0.000000m
Iteration 12/1101: speed=5km/h camera=34ms, odom(quality=0.844877, kfs=1)=133ms, slam=0ms, rmse=0.000000m
Iteration 13/1101: speed=6km/h camera=27ms, odom(quality=0.841638, kfs=1)=137ms, slam=0ms, rmse=0.000000m

And it goes on until it prints this:

Iteration 1100/1101: speed=0km/h camera=18ms, odom(quality=0.881621, kfs=63)=197ms, slam=0ms, rmse=0.000000m
Iteration 1101/1101: speed=0km/h camera=18ms, odom(quality=0.879493, kfs=63)=193ms, slam=8ms, rmse=0.000000m
Total time=214.098361s
Saving trajectory ...
Saving /home/antonis/07/rtabmap_poses.txt... failed!
Ground truth comparison:
   KITTI t_err = -nan %
   KITTI r_err = -nan deg/m
   translational_rmse=   0.000000 m
   rotational_rmse=      0.000000 deg
Saving rtabmap database (with all statistics) to "/home/antonis/07/rtabmap.db"
Do:
 $ rtabmap-databaseViewer /home/antonis/07/rtabmap.db

Any ideas about what I do wrong ?

[matlabbe]

Hi Mikor,

Remove "--RGBD/Enabled false\". I didn't realize there was no description for this parameter, which I assumed it should be always true for everyone. I added a description in this commit.

cheers,
Mathieu

[Mikor]

Okay now it works great, thank you again!

I can see that in the database I get around 200 poses but the groundtruth data has around 1100. How does this difference occur ?

Also, what do the poses of the txt file mean? I can see that there are 12 elements in every pose, are they referring to a translation and rotation matrix meaning that the first 3 elements describe the translations and the other 9 are the elements of the rotation matrix ?

Regards,
Anthony.

[matlabbe]

Hi Anthony,

The detection rate is 2 Hz, and data is 10 Hz, so there is one pose added to map each 5 scans. To have all poses, you can add "--Rtabmap/CreateIntermediateNodes true".

For the kitti 12 values format, it is the 3x4 transformation matrix if I remember correctly:
r00,r01,r02,t0,r10,r10,r11,r12,t1,r20,r21,r22,t2

cheers,
Mathieu

[Mikor]

Hey Mathieu,

thanks again for your response, but it seems that I am not getting an exact match of poses. I work on this example in the 07 sequence (because it is smaller and it saves time), when I use the /Rtabmap/CreateIntermediateNodes I get a few less poses than the groundtruth txt contains 1101. I thought that when I change the  My second thought based on this post  was that it doesn't create a crucial key frame in order to fill the intermediate nodes and I set the threshold to lower values but this does not seem to work and the errors increase so even it worked it wouldn't be ideal.

With detection rate set at 2 and the use of intermediate nodes I get 1020 poses.
Same detection rate and without the intermediate nodes I get 208 poses.
When I set the detection rate at 10 and with or without intermediate nodes I get 1018 poses.

If I open the database it says that 1101 ids loaded but if I export map's graph it results in 1020 poses and if export odometry then the poses match the 1101, should I do the latter and do the rest comparisons to test my parameters??

Also you are correct about the format of kitti values. I double checked the text file in devkit that KITTI provides and it says the same thing.

With gratitude,
Anthony

[matlabbe]

Ah yeah, you have to add also "--RGBD/LinearUpdate 0". Without RGBD/LinearUpdate disabled, when the car is not moving, no nodes are added to graph. It is counter-intuitive to add nodes while not moving, but to comply to dataset, we have to add them. I get 1020 nodes when enabled, 1101 when disabled.

cheers,
Mathieu

[felipeduque]

Unscrupulously hijacking this thread: I'm facing the issue of smaller amount of poses obtained than the ground truth for KITTI, when using "rtabmap-kitti_dataset" with images and Velodyne.

In sequence 0, there are 4541 pictures, and I can see all 4541 nodes in the database-viewer when the detection rate is 2Hz and "createIntermediaryNodes=true". However, the 3D map shows very sparse point clouds, probably because the intermediate nodes were not assigned their point clouds because of the 2Hz rate.

I really need all nodes to have their point clouds, so I set the rate to 10Hz, but to my surprise, now I had only 384 nodes, even with the intermediary nodes option being set. The same amount of nodes is obtained when the rate is set to 20Hz. What could be going on? All the other options for "rtabmap-kitti_dataset" are the standard ones shown in the help text, with "--scan".

EDIT: "rtabmap-report" says that odometry processing time stays below 100ms, so probably it's not a CPU processing issue. (However, my RTAB-Map version is 0.17.6, the statistics is called "Odometry/TotalTime/ms", but hopefully it means the same thing.)

EDIT 2: I solved it partially by realizing that sequence 00 had several straight line segments, and "RGBD/AngularUpdate = 0.1" (its default value) prevented new nodes from being generated. After setting it to zero, I finally obtained all nodes matching with the ground truth, even without creating intermediary ones. However, several nodes do not have depth data ("SensorData.depthRaw().empty()" is true for them). Why is that?

[felipeduque]

Solved. I was treating Velodyne data as point cloud, since it's in 3D, but it's a regular scan data in RTAB-Map. So I had to use stereo option ("-- disp" argument in "rtabmap-kitti_dataset") to build the point cloud. All good now!