use_backend=true fails on KITTI & OMD datasets (RGB-D mode)

[sally-203]

Dear author,

Thanks for your open-sourced code!

I am trying to run the system with use_backend=true, but it crashes consistently on both KITTI and OMD datasets. I did not modify any code or parameters. However, when I set use_backend=false, the system runs normally without any errors. The error screenshot is attached below.

‘ ros2 run dynosam_utils eval_launch.py --output_path=/tmp/results --name=test --run_pipeline --data_provider_type=3 --dataset_path=/root/data/omd/’

'ros2 launch dynosam_ros dyno_sam_launch.py params_path:=/home/user/dev_ws/src/core/dynosam/params/ dataset_path:=/root/data/omd/ output_path:=/root/results/omd v:=0'

Do I need to adjust specific parameters in the YAML config file for different datasets when enabling use_backend? Are there recommended parameters for KITTI or OMD? If not a code bug, could you provide a reference configuration (params) that is known to work with these datasets?

[jessemorris]

Hi, Yes it has come to my attention that there are a few issues like this, mostly a result of me doing development in parallel! I am currently in the processing of pushing fixes to the main/develop branches with a ~1-2 week timeline. In the meantime, please try running the code using one of the test scripts in dynosam_utils (ie. run_experiments_ecmr.py, which is actually the results for our RA-L paper).

[MarceloContreras]

I came to notice that in order to get similar results to the T-RO paper, it is better to utilize the settings from the previously mentioned script: run_experiments_tro.py from dynosam_utils package. Overall those settings are: Full Batch optimization (incremental always crashes and moving window seems not been implemented) and either Motion or Pose estimator factors.