🚘 This is the official GitHub repository of the paper:

OpenVO: Open-World Visual Odometry with Temporal Dynamics Awareness
Phuc Nguyen, Anh Nhu, Ming Lin
CVPR 2026

OpenVO, a novel framework for Open-world Visual Odometry (VO) under limited input conditions.

Details of the model architecture and experimental results can be found in our paper

@article{nguyen2026openvo,
  title={OpenVO: Open-World Visual Odometry with Temporal Dynamics Awareness},
  author={Nguyen, Phuc DA and Nhu, Anh N and Lin, Ming C},
  journal={arXiv preprint arXiv:2602.19035},
  year={2026}
}

Please CITE our paper whenever this repository is used to help produce published results or incorporated into other software.

• State-of-the-art performance on metric-scale zero-shot visual odometry!
• Support KITTI, nuScenes, Argoverse2 and many MORE!
• Reproducibility code for the community!
• Continuous maintenance/improvement for stronger version!

Please refer to installation guide

Please refer to data preparation

Adjust the paths accordingly in my_inference.py:

root = './data'
weights = './weights/OpenVO/openvo_nusc_gt'
save_path = "./results"

Run

python src/my_inference.py --config configs/openvo.py

If you have multiple GPUs, toggle nprocess=$NUM_GPUs$ and devices=[0,1,2,...$NUM_GPUs$-1] in main(). This would launch multiprocess with shared CPU/RAM for mass inference.

For visualization, toggle the visualizer on.

Run

cd ./odom-eval
python eval.py

The results will be saved under ./odom-eval/evaluation_results

Adjust the paths in eval.py accordingly:

eval_dirs = [
            "openvo_nusc_gt"
            ]
result_dir_openvo = '../results'

Note on Argoverse2_Stereo only: All results reported in the paper were obtained using models trained with ground-truth intrinsics from stereo videos. The corresponding numerical results are available in ./odom-eval/evaluation_results.

ARGO2_Stereo: 12.59, 2.63, 7.71, 0.13 KITTI_X: 9.06, 3.45, 96.24, 0.06 NUSC_X: 9.04, 3.86, 5.90, 0.10

After preparing the data, train the model.

Single GPU training:

python src/main.py --config configs/openvo.py

Check your log while training...(there would be no on-screen display)

Please keep an eye out for the regular update!