search

PRBonn / semantic_suma

SuMa++: Efficient LiDAR-based Semantic SLAM (Chen et al IROS 2019)
MIT License
923 stars 205 forks source link

the effect of semantic segmentation #26

Closed ACFFF closed 4 years ago

ACFFF commented 4 years ago

Hi，when i run semantic_suma on my computer，the effect of semantic segmentation is not the same as the your demo，This is the effect I run: image compare with yours: image

i used the darknet53.

jbehley commented 4 years ago

Please see the solutions provided in https://github.com/PRBonn/rangenet_lib/issues/9 related to Nvidia's RTX Series which are as follows:

1.The rightcombination of CUDA and TensorRT.

  1. Others had success with using another ONNX version.
  2. Finally, a change in the codecould resolve it: https://github.com/PRBonn/rangenet_lib/issues/9#issuecomment-644628222

Hope one of the described solutions works.

ACFFF commented 4 years ago

Please see the solutions provided in PRBonn/rangenet_lib#9 related to Nvidia's RTX Series which are as follows:

1.The rightcombination of CUDA and TensorRT.

  1. Others had success with using another ONNX version.
  2. Finally, a change in the codecould resolve it: PRBonn/rangenet_lib#9 (comment)

Hope one of the described solutions works.

thank you.i used the third way to slove it.the model darknet53 can be useful in kitti velodyne while it doesnt work in my data.i will learn https://github.com/PRBonn/lidar-bonnetal to train a model that fits my own data.

Chen-Xieyuanli commented 4 years ago

Hey @ACFFF, since the problem was solved, I'm going to close this issue.

If there is any further problem, please re-open it again.