This repo implements the network described in the ICRA2022 paper:

Self-Supervised Ego-Motion Estimation Based on Multi-Layer Fusion of RGB and Inferred Depth

Zijie Jiang, Hajime Taira, Naoyuki Miyashita and Masatoshi Okutomi

Please also visit our project page.

If you find our work useful for your research, please consider citing the following paper:

@inproceedings{jiang2022mlfvo,
  title={Self-Supervised Ego-Motion Estimation Based on Multi-Layer Fusion of RGB and Inferred Depth},
  author={Jiang, Zijie and Taira, Hajime and Miyashita, Naoyuki and Okutomi, Masatoshi},
  booktitle={2022 IEEE International Conference on Robotics and Automation (ICRA)},
  year={2022}
}

• 2022.05.20. Pretrained weights and demo for testing are open.

Our experiments are conducted on a machine installed with Ubuntu 18.04, Pytorch 1.7.1, CUDA 10.1. You can install other required packages by:

pip install requirements.txt

Since our demo partly depends on Monodepth2 and CEN, clone this repository with --recurse option. Please also refer their original repositories.

Please download the KITTI odometry benchmark from their site and create a soft link by executing the following command:

ln -s /path/to/dataset/data_odometry_color/dataset data

Download and place the pretrained weights of models following:

wget http://www.ok.sc.e.titech.ac.jp/res/MLF-VO/models.zip
unzip -d data models.zip
rm models.zip

The demo specifies several paths and configurations in configs/run_odometry.yaml. Please modify line 2, 6, or 9 to use your own paths to dataset, model weights, or outputs.

python run_odometry -c configs/run_odometry.yaml

The estimated trajectory will be exported to data/outputs/09.txt. Each line in the file represents the estimated camera, corresponding to each frame in the sequence, in the following format:

T11 T12 T13 T14 T21 T22 T23 T24 T31 T32 T33 T34
...

where T11, T12, ... , T34 are the elements of 3x4 camera transformation matrix (from camera to world coordinate system):

T11 T12 T13 T14
T21 T22 T23 T24
T31 T32 T33 T34

We recommend to use this toolbox to evaluate the inferred trajectory.

Please checkout to the dev branch, and run

python train.py --data_path ~/Documents/datasets/kitti_odom/dataset --log_dir ~/Documents/MLF-VO/KITTI --model_name test --num_epochs 20 --dataset kitti_odom --split odom

Please change L55 in trainer.py to the wanted type of the PoseNet.

self.models['pose'] = posenet_type_dict[config_dict.model.type]()

We are grateful to the authors of Monodepth2 and CEN for publicly sharing their codes.

MLF-VO is released under GPLv3 License.