search

htkseason / VOLDOR

VOLDOR-SLAM is a real-time dense-indirect SLAM system takes dense optical flows as input that supports monocular, stereo and RGB-D video sequence.
472 stars 56 forks source link

Reproduce VO performance #9

Closed jinyummiao closed 3 years ago

jinyummiao commented 3 years ago

Hi, thanks for your work. I have successfully built VO pipeline according to your instructions. However, it fails to localize or map when I run VO pipeline in monocular mode. The output log may be helpful to fix this problem:

GPUassert : invalid device symbol ../../gpu-kernels/collect_p3p_instances.cu at line 186 GPUassert : no kernel image is available for execution on the device ../../gpu-kernels/reduce_vector_sum.h at line 53 Tracking lost at 6

Looking forward to your reply:)

jinyummiao commented 3 years ago

The command I used: python demo.py --fx 360.76885 --fy 360.76885 --cx 304.77965 --cy 86.427 --bf 192.19074 --flow_dir '../KITTI-03/flow_mfn' --img_dir '../KITTI-03/img' --mode mono --resize 0.5 --abs_resize 0.25

htkseason commented 3 years ago

Hi there, It seems a compling problem. Could you share with me your OS, GPU model and CUDA version? What is the nvcc_machine_code you are using in your 'setup.py'?

jinyummiao commented 3 years ago

I compiled the project on a laptop with GTX 1660Ti GPU, CUDA 10.0, Ubuntu 18.04. I have tried the original nvcc_machine_code='-m64 -arch=compute_61 -code=sm_61', ncc_machine_code='' and nvcc_machine_code='-m64 -arch=compute_75 -code=sm_75' (referring to this). All failed...

jinyummiao commented 3 years ago

It is concluded as a compling problem. I have fixed the problem. Thanks for your kindly help!