If you find this useful, please cite the related paper:
@article{DBLP:journals/corr/abs-1906-04838,
author = {Kevin Christensen and
Martial Hebert},
title = {Edge-Direct Visual Odometry},
journal = {CoRR},
volume = {abs/1906.04838},
year = {2019},
url = {http://arxiv.org/abs/1906.04838},
archivePrefix = {arXiv},
eprint = {1906.04838},
}
This repository assumes the following directory structure, and is setup for the TUM-RGBD Dataset:
EdgeDirectVO
|-- tum_rgbd_dataset/
|-- depth/
|-- rgb/
|-- assoc.txt
|-- tum_rgbd_dataset_ground_truth.txt
Be sure to run assoc.py to associate timestamps with corresponding frames.
cd build && cmake .. && make -j && ./EdgeDirectVO
After evaluating on a dataset, the corresponding evaluation commands will be printed to terminal. Simpy copy and run them in terminal in project root directory. Examples are shown below:
For relative pose error:
python evaluate_rpe.py groundtruth_fr1xyz.txt rgbd_dataset_freiburg1_xyz_results.txt --fixed_delta --delta 1 --verbose
python evaluate_ate.py groundtruth_fr1xyz.txt rgbd_dataset_freiburg1_xyz_results.txt --verbose