search

mchancan / deepseqslam

The Official Deep Learning Framework for Robot Place Learning
https://mchancan.github.io/spl
GNU General Public License v3.0
92 stars 14 forks source link

can relative poses be used as positional information? #5

Open guyupan1911 opened 12 months ago

guyupan1911 commented 12 months ago

Hi, thank you for you work. It seems that the same GPS data is used in both train and test procedure. In the paper says other position sources such as odometry can be used. should i use relative poses or global poses of two different trips?

mchancan commented 10 months ago

Hi - sorry for the late reply and thank you for your interest in this work. You can use relative poses if training and testing trajectories have roughly the same start (e.g. point A) and end (e.g. point B) locations; this is typically the case for autonomous driving datasets such as the Oxford RobotCar dataset for instance. Otherwise you'd need to transform relative poses to global for positional consistency.