search

xaldyz / dataflow-orbslam

Real-Time SLAM for Monocular, Stereo and RGB-D Cameras, with Loop Detection and Relocalization Capabilities
Other
94 stars 12 forks source link

About stereo and rgb-d node #5

Open seomk97 opened 2 years ago

seomk97 commented 2 years ago

Hello, Can this code run on stereo or rgb-d node? Is there any test result for stereo and rgb-d node?

Thank you!!

seomk97 commented 2 years ago

I modified code to use OpenVX GPU for RGB-D Luckily, RGB-D mode uses same ORBextractor as Monocular mode so little work needed. But pipelining option incurred seg fault error while operating the code so I disabled and used only OpenVX gpu option. I got around 100 ms per frame(TUM RGB-D frei1/desk1 dataset) for this modified version and 130ms for all the optimization option disabled version. I am happy to get 30ms fast result. but I have problem that both these resuts are slower than official ORB-SLAM2 result(~80ms). Both monocular and RGB-D mode with no optimization option in this repo is slower than original implementation. Although I disabled all optimization options in cmake, there is something different with original implementation of ORB-SLAM2 I guess. Can I get the reason of this?