Monocular Per-Object Distance Estimation with Masked Object Modeling

This repository contains the officila implementation for our paper:

Monocular Per-Object Distance Estimation with Masked Object Modeling
arXiv:2401.03191

🚧 Work in Progress 🚧
This project is still under development. Expect updates and improvements.

Overview

Dataset Preparation

Download and prepare the datasets following the instructions in DATASET.md (to be added).

Running the Code

To run experiments with the best hyperparameters for each dataset, use the provided scripts:

bash nuscenes.sh  # Run on NuScenes dataset
bash motsynth.sh  # Run on MOTSynth dataset
bash kitti.sh     # Run on KITTI dataset

Each script is pre-configured with optimal hyperparameters for the respective dataset.

Citation

If you find this work useful, please cite our paper:

@article{panariello2025monocular,
  title={Monocular Per-Object Distance Estimation with Masked Object Modeling},
  author={Panariello, Aniello and Mancusi, Gianluca and Haj Ali, Fedy and Porrello, Angelo and Calderara, Simone and Cucchiara, Rita},
  journal={Computer Vision and Image Understanding},
  year=2025
}

Contact

For questions or collaborations, please open an issue or reach out via email.