openxrlab / xrdslam

Platform for Deep Learning based SLAM
Apache License 2.0
187 stars 21 forks source link

The program is stuck with no error reported #9

Closed DY-ATL closed 6 days ago

DY-ATL commented 1 week ago

The terminal looks like this

[19:20:20] Saving config to: outputs/config.yml
  0%|                                                      | 0/2000 [00:00<?, ?it/s]
wangxiaomeng030 commented 1 week ago

Which algorithm is being run here? Some algorithms iterate 1000 times on the first frame, so they may run slowly. You can try Co-SLAM first to quickly verify the environment setup; alternatively, you can directly modify the iteration count for the corresponding algorithm in input_config.py to avoid it getting stuck.

DY-ATL commented 1 week ago

I used point-slam and stucks. I try co-slam and it works! Thank you!

Is the speed normal for a RTX3090 GPU? 64%|███████████████▎ | 1285/2000 [22:00<12:58, 1.09s/it]

I'm new to SLAM area. Can you recommend some paper on dense visual SLAM? (We have monocular/stereo camera, IMU, without depth map. Deep learning based methods are preferred. Need to run real-time on robot.) Thanks again!

wangxiaomeng030 commented 6 days ago

Co-SLAM runs very quickly, while Point-SLAM and SplaTAM run relatively slower. You can refer to the runtime details in our paper. Additionally, the algorithms currently integrated in XRDSLAM only support RGBD-SLAM. If you don’t have depth images, you may consider using an RGB-SLAM approach, such as Splat-SLAM.

DY-ATL commented 6 days ago

Thank you!