The repository contains the code and data for "Towards Robust Human Trajectory Prediction in Raw Videos" IROS 2021.
Download the Stanford Drone Dataset (SDD). The ~69G zip file includes the raw videos and annotations.
Our refined annotations of SDD 📢
Download the WILDTRACK Seven-Camera HD Dataset.
Our extended annotations of WILDTRACK 📢
The project was developed based on Python 3.7.9 and PyTorch 1.1.0. The environment can be set up via the environment.yml file:
conda env create -f environment.yml
conda activate retracking
As an example, we show how to run the experiments on the SDD dataset.
Train a prediction model
cd retracking-by-prediction
python train.py
Detection, Tracking, Prediction, and Re-tracking
sh run_test.sh
Evaluation
We evaluate the tracking performance via py-motmetrics. Note: we use L2 distance in meter instead of box IoU as the association metric.
If you use the code or data in your research, please cite the paper:
@inproceedings{Yu2021-Retracking,
author = {Rui Yu and Zihan Zhou},
title = {Towards Robust Human Trajectory Prediction in Raw Videos},
booktitle = {{IEEE/RSJ} International Conference on Intelligent Robots and Systems, {IROS}},
year = {2021}
}
The project is released under the MIT License. The SORT tracking code should follow its own license (GPL-3.0).