Evaluate

[yasrim1990]

Hi mathieu,
I have some question in used your great rtabmap ..
Please help me..

How can I evaluate the rtabmap resaults like as RMSE and time passed with other slam solver ways?
Do you have any evaluation of rtabmap with other ways to show me the degree of rtabmap's goodness?
( because you didn't have published any paper about that )

For example i open a new dataset like TUM's datasets , can i see the RMSE and the time passed to create full odometry and mapping?

I already had been saw , rtabmap could count the feature's extracted... which window show that?

In your idea, which discriptor is better ( surf , sift  or ORB )? And how rtabmap can show that?

Thanks

[m.omar82]

HI I use Rtabmap, there is tool call "rtabmap-console" it is in the bin directory, I use to test rtabamp to get evaluation there is two steps: 1. this is example how to use rtabmap-console with Lib6Indoor dataset Benchmark lip6-indoor there will be two output files (LogF.txt and LogI.txt) to the next step. 2. rtapmap have Matlab/ShowLogs folder have tool to evaluation the dataset showlogs.m and you need to download the ground truth Lip6Indoor.png : showlogs('/home/mohammed/Document/Rtab-map.', '/home/mohammed/Document/Lip6Indoor.png'); this will show you full information (time memory evaluation ) about your run more information can see in this pager --------------------------------------------------------- for SIFT ORB use please this page

[matlabbe]

For 3D SLAM, there is no benchmark yet...

EDIT To see the number of features extracted, open the Statistics view (Window->Show View->Statistics), then when rtabmap is running you should see some panels with many statistics. You can have the number of features extracted from Odometry under Odometry tab, and features extracted for loop closure detection under Keypoint tab.

My thoughts are that binary features are better for odometry (because they are faster to extract, so faster odometry) but float descriptors like SURF/SIFT would be better for loop closure detection.

[yasrim1990]

thanks Dear Mohammed

[yasrim1990]

Thanks my friend ..