mingyuyng
Training on a different Dataste
I am trying to train this Neural Network on a different dataset but that is not working, I have tracked this down to thew creation of the relative poses and their conversion back into absolute poses, This operation is being done well and correctly for the KITTI dataset but for other datasets like the EuRoC and my own dataset form simulation is not correct! when I take the quaternions and positions from my dataset and convert them into the R(rotation) | t (position) 4 by 4 matrices like for the KITTI dataset and then convert them into poses_rel like in the code and then back into absolute poses using path_accu, the absolute poses I get are not the same. What should I do in this case ? help would be appreciated!
