search

kajo-kurisu / D_VINS

Merge superpoint、lightglue、MixVPR into VINS-FUSION for loop closure with TensorRT
GNU General Public License v3.0
62 stars 8 forks source link

savePoseGraph? #8

Open jingyang-huang opened 3 months ago

jingyang-huang commented 3 months ago

Hi, D_VINS is a marvellous work which combines network and VIO. Thanks to your guidance, I have run this project on my own dataset, and the loop detection works pretty well. However, when I try to save the pose graph of my own dataset, the process seems to be stuck and just saves a few keyframes with very big descriptors, and it does not save any keyframes anymore. image img_v3_02cj_f021b846-311b-4f68-b239-aae925e329ag To figure out if this is caused by my own dataset, I conducted the same step in euroc dataset. But the situation is different here: loop detection works well, but the ROS loop fusion process died, and it seems no descriptors are saved. img_v3_02cj_f399179e-42af-4a00-a3eb-4c1b4c1cb67g Waiting for your reply and thanks again~

kajo-kurisu commented 3 months ago

Thank you for your attention. Due to the many different methods used in the project, the ability to save posegraph is still being refined and the the function is not available yet!

kajo-kurisu commented 3 months ago

If you want to perform multi-session experiments, just play the second bag after the first one is finished. Our approach supports this function.

jingyang-huang commented 3 months ago

If you want to perform multi-session experiments, just play the second bag after the first one is finished. Our approach supports this function.

Thank you for your advice, the multi-session experiment works well. Good luck to your paper and looking forward to a completed version!