eval-vislam

Toolkit for VSLAM and VISLAM evaluation.

For more information, please refer to our project website.

License

This project is released under the Apache 2.0 license.

Prerequisites

ISMAR 2019 SLAM Challenge Scoring ToolKit

Usage:
  python3 ismar-score.py --round <round> --is_vislam <is_vislam> --trajectory_base_dir <trajectory_base_dir> --gt_base_dir <gt_base_dir>

Arguments:
  <round>                    SLAM benchmark number of rounds, e.g. 5.
  <is_vislam>                Set to 1 for VISLAM, set to 0 for VSLAM.
  <trajectory_base_dir>      SLAM camera trajectory file folders(e.g. ~/MY-SLAM/trajectories/). We support TUM format(timestamp[s] px py pz qx qy qz qw) files.
  <gt_base_dir>              Path to groundtruth folder, e.g. ~/ISMAR-Dataset/train.

Toolkit Usage

Accuracy (APE, RPE, ARE, RRE, Completeness)

Usage:
  ./accuracy <groundtruth> <input> <fix scale>

Arguments:
  <groundtruth>    Path to sequence folder, e.g. ~/VISLAM-Dataset/A0.
  <input>          SLAM camera trajectory file in TUM format(timestamp[s] px py pz qx qy qz qw).
  <fix scale>      Set to 1 for VISLAM, set to 0 for VSLAM.

Initialization Scale Error and Time

Usage:
  ./initialization <groundtruth> <input> <has inertial>

Arguments:
  <groundtruth>    Path to sequence folder, e.g. ~/VISLAM-Dataset/A0.
  <input>          SLAM camera trajectory file in TUM format(timestamp[s] px py pz qx qy qz qw).
  <has inertial>   Set to 1 for VISLAM, set to 0 for VSLAM.

Robustness

Usage:
  ./robustness <groundtruth> <input> <fix scale>

Arguments:
  <groundtruth>    Path to sequence folder, e.g. ~/VISLAM-Dataset/A0.
  <input>          SLAM camera trajectory file in TUM format(timestamp[s] px py pz qx qy qz qw).
  <fix scale>      Set to 1 for VISLAM, set to 0 for VSLAM.

Relocalization Time

Usage:
  relocalization <groundtruth> <input> <has inertial> <jump detection>

Arguments:
  <groundtruth>    Path to sequence folder, e.g. ~/VISLAM-Dataset/A0.
  <input>          SLAM camera trajectory file in TUM format(timestamp[s] px py pz qx qy qz qw).
  <has inertial>   Set to 1 for VISLAM, set to 0 for VSLAM.
  <jump detection> Sensitivity to detect jump when relocalization happened (Default 0.05).

Citation

If you are using our codebase or dataset for research, please cite the following publication:

@article{
  title   = {Survey and Evaluation of Monocular Visual-Inertial SLAM Algorithms for Augmented Reality},
  author  = {Jinyu Li, Bangbang Yang, Danpeng Chen, Nan Wang, Guofeng Zhang*, Hujun Bao*},
  journal = {Journal of Virtual Reality & Intelligent Hardware},
  year    = {2019},
  volume  = {1},
  number  = {4},
  pages   = {386-410},
  url     = {http://www.vr-ih.com/vrih/html/EN/10.3724/SP.J.2096-5796.2018.0011/article.html},
  doi     = {10.3724/SP.J.2096-5796.2018.0011}
}

EyeeHKing / eval-vislam