search

LuoXubo / JointLoc

[IROS 2024] JointLoc: A Real-time Visual Localization Framework for Planetary UAVs Based on Joint Relative and Absolute Pose Estimation
https://luoxubo.github.io/projects/iros2024.html
MIT License
35 stars 1 forks source link

The result about LoFTR in Table V #11

Open RingoWRW opened 1 month ago

RingoWRW commented 1 month ago

Hi Luo, I believe JointLoc is a remarkable achievement and I am intrigued by this work. I have a question about Table V. Why can't LoFTR generate matches in textureless regions?

LuoXubo commented 1 month ago

Hi, sorry for the confusion. The "failed" here refers to the fact that our device cannot support the matching of LoFTR in large-scale map localization.

RingoWRW commented 1 month ago

Thank you for your response. I also have another question regarding Table V. I would like to know the experimental conditions for absolute localization. I noticed that the error for absolute localization is 10 times larger than SLAM.

LuoXubo commented 1 month ago

The experiments are conducted on the RTX 2080Ti. Absolute localization sometimes has very small errors, and sometimes has very large errors. I think the main reason why absolute localization cannot be used alone for UAV is that the positioning result is not smooth enough.