fangget / lsd_slam_no_ros_pangolin Goto Github PK

Scanning dependencies of target lsd_slam_core
[ 3%] Building CXX object CMakeFiles/lsd_slam_core.dir/src/GlobalMapping/KeyFrameGraph.cpp.o
[ 7%] Linking CXX shared library ../lib/liblsd_slam_core.so
/usr/bin/ld: can not found -lcsparse
collect2: error: ld returned 1 exit status
CMakeFiles/lsd_slam_core.dir/build.make:413: recipe for target '../lib/liblsd_slam_core.so' failed
make[3]: *** [../lib/liblsd_slam_core.so] Error 1
CMakeFiles/Makefile2:109: recipe for target 'CMakeFiles/lsd_slam_core.dir/all' failed
make[2]: *** [CMakeFiles/lsd_slam_core.dir/all] Error 2
CMakeFiles/Makefile2:84: recipe for target 'CMakeFiles/dataset_lsd_slam.dir/rule' failed
make[1]: *** [CMakeFiles/dataset_lsd_slam.dir/rule] Error 2
Makefile:118: recipe for target 'dataset_lsd_slam' failed
make: *** [dataset_lsd_slam] Error 2

Hi, FangGet

I have a problem that tracking is lost after several frames when running your code and testing tum and EuRoC dataset.
tum(rgbd_dataset_freiburg2_desk):

0.813920 1.085432 0.508034 0.520212 0 640 480 crop 640 480
EuRoC(V1_01_easy):

0.609912 0.952700 0.488318 0.517448 0 752 480 full 752 480
they all lost tracking, and below it's camera calibration.

Do you have any idea about this? Except viewer, do you change any code so that this problem occurs?

Hi,

I have already compiled your LSD-SLAM version with Pangolin. It is working but i am getting bad reconstructions. What I mean with "bad" is that the pointcloud contains too much noise and many points seem to be bad triangulated (forming the typical cone of points). My results are far away from the original videos.

I am working with an HP Webcam at 640x480. It is calibrated with the FOV-model correctly (PTAM model).

Are you getting the same results? Otherwise, could you tell me how to improve them?

Thanks in advance,

Jon

Can you tell me how you test the dataset of KITTI or TUM?