Pose plots using evo look different from the paper about sc-sfmlearner-release HOT 5 OPEN

Hi Sai, the reason is because in this work, we were using the released pose result (original paper) of SfM-Learner instead of running the model (updated).
You can find the released result from here:
https://github.com/tinghuiz/SfMLearner/blob/master/kitti_eval/download_kitti_pose_eval_data.sh

For the comparison between the updated SfM-Learner, you can see it from our latest work.
https://arxiv.org/abs/1909.09803

If you try to visualize the old result of sfmlearner using evo, you will find that there is a rotation issue in seq10 and it is easier to be visualized in 3D instead of the weird 2D visualization result in the figure.

from sc-sfmlearner-release.

Btw, the SfMLearner result contains 5-frame snippets instead of full trajectory. In order to convert the snippets to a full trajectory, you may need this tool as well.
https://github.com/Huangying-Zhan/Depth-VO-Feat/tree/master/tools/sfmlearner_odometry_tool

from sc-sfmlearner-release.

Hi Sai, Can you share the code of the visual SfM-Learner's pose?

from sc-sfmlearner-release.

@SaiK95 Hi Sai, Can you share the code of the visual SfM-Learner's pose?

from sc-sfmlearner-release.

@Huangying-Zhan @Samyer Hi, isit possible for you to share the predictions.npy of the sequence 10 of the SFM-Learner that is shown in the SC-SFM Learner Paper? The one with the rotational error before it got updated?
Thanks in advance

from sc-sfmlearner-release.