Unified Human Localization and Trajectory Prediction with Monocular Vision

Po-Chien Luan*, Yang Gao*, Celine Demonsant and Alexandre Alahi

IEEE International Conference on Robotics & Automation (ICRA) 2025

[Paper]

Getting Started

Create a new conda environment:

conda create -n mt python=3.9
conda activate mt

Install PyTorch from here accoding to your environment.

Install the requirements using pip:

pip install -r requirements.txt

Install PyTorch3D from here according to your environment. Take Python=3.9, CUDA=11.8, PyTorch=2.4.1 as an example:

conda install https://anaconda.org/pytorch3d/pytorch3d/0.7.8/download/linux-64/pytorch3d-0.7.8-py39_cu118_pyt241.tar.bz2

Data

Download the preprocessing data from our release and unzip the data. We will release the preprocessing code soon.

Training and Testing

Train pedestrian localization:

python train.py --train_mode loc 

Freeze pedestrian localization and train trajectory prediction:

python train.py --train_mode freeze_loc

Train two modules jointly:

python train.py --train_mode joint

Evaluate localization:

python eval.py --eval_mode loc 

Evaluate trajectory prediction:

python eval.py --eval_mode traj_pred 

You need to set the following arguments correctly:

• joints folder: path to your data.
• load_loc: path to your localization checkpoint.
• load_traj: path to your trajectory prediction checkpoint.
• loc_cfg: path to your configuration of localization model.
• traj_cfg: path to your configuration of trajectory prediction model.

Please check test_script.sh for some examples. We also provide checkpoints in release to reproduce the main results in the paper.

For citation

This repository is work-in-progress and will continue to get updated and improved over the coming months.

@article{luan2025unified,
  title={Unified Human Localization and Trajectory Prediction with Monocular Vision},
  author={Luan, Po-Chien and Gao, Yang and Demonsant, Celine and Alahi, Alexandre},
  journal={arXiv preprint arXiv:2503.03535},
  year={2025}
}

This work is a follow-up work of MonoLoco and Social-Transmotion.

@InProceedings{saadatnejad2024socialtransmotion,
    title={Social-Transmotion: Promptable Human Trajectory Prediction}, 
    author={Saadatnejad, Saeed and Gao, Yang and Messaoud, Kaouther and Alahi, Alexandre},
    booktitle={International Conference on Learning Representations (ICLR)},
    year={2024}
}

@InProceedings{bertoni_2019_iccv,
    author = {Bertoni, Lorenzo and Kreiss, Sven and Alahi, Alexandre},
    title = {MonoLoco: Monocular 3D Pedestrian Localization and Uncertainty Estimation},
    booktitle = {the IEEE International Conference on Computer Vision (ICCV)},
    month = {October},
    year = {2019}
}