NOTE: This repository contains most of the original implementation for processing datasets to be in Trajnet++ format. Apart from some codestyle changes and minor bugfixes, the major difference here is removal of usage of rvo2 libary This libary was being used for synthetic data generation and processing, but I was unable to use it in Windows. As such, while this version works on windows - following the tutorial below, apart from installing rvo2 libary - you cannot convert the ORCA synthetic files to use in Trajnet++ formats. You can get the train and test files (no validation nor ground truth) here
NEW: Converting new external dataset into TrajNet++ format. Tutorial
pip install -e '.[test,plot]'
pylint trajnetdataset
pytestExisting real world data:
data/
data_arxiepiskopi.rar
data_university_students.rar
data_zara.rar
ewap_dataset_light.tgz
# 3DMOT2015Labels # from: https://motchallenge.net/data/3DMOT2015Labels.zip (video file at http://cs.binghamton.edu/~mrldata/public/PETS2009/S2_L1.tar.bz2)
Train.zip # from trajnet.epfl.ch
cvpr2015_pedestrianWalkingPathDataset.rar # from http://www.ee.cuhk.edu.hk/~syi/ (website not accessible but data are also here: https://www.dropbox.com/s/7y90xsxq0l0yv8d/cvpr2015_pedestrianWalkingPathDataset.rar?dl=0.+63)
cff_dataset.zip # from https://www.dropbox.com/s/cnnk2ofreeoshuz/cff_dataset.zip?dl=0
Extract:
# biwi
mkdir -p data/raw/biwi
tar -xzf data/ewap_dataset_light.tgz --strip-components=1 -C data/raw/biwi
# crowds
mkdir -p data/raw/crowds
unrar e data/data_arxiepiskopi.rar data/raw/crowds
unrar e data/data_university_students.rar data/raw/crowds
unrar e data/data_zara.rar data/raw/crowds
# cff
mkdir -p data/raw/cff_dataset
unzip data/cff_dataset.zip -d data/raw/
rm -r data/raw/__MACOSX
# Wildtrack: https://www.epfl.ch/labs/cvlab/data/data-wildtrack/
mkdir -p data/raw/wildtrack
unzip data/Wildtrack_dataset_full.zip -d data/raw/wildtrack
# L-CAS: https://drive.google.com/drive/folders/1CPV9XeJsZzvtTxPQ9u1ppLGs_29e-XdQ
mkdir -p data/raw/lcas
cp data/lcas_pedestrian_dataset/minerva/train/data.csv data/raw/lcas
# pedestrian walking dataset
mkdir -p data/raw/syi
unrar e data/cvpr2015_pedestrianWalkingPathDataset.rar data/raw/syi
PETS09 S2L1 ground truth -- not used because people behavior is not normal
mkdir -p data/raw/mot
unzip data/3DMOT2015Labels.zip -d data/
cp data/3DMOT2015Labels/train/PETS09-S2L1/gt/gt.txt data/raw/mot/pets2009_s2l1.txt
# Edinburgh Informatics Forum tracker -- not used because tracks are not good enough
mkdir -p data/raw/edinburgh
wget -i edinburgh_informatics_forum_urls.txt -P data/raw/edinburgh/Prepare synthetic data:
python -m trajnetdataset.controlled_dataHelp menu for generating diverse synthetic data:
python -m trajnetdataset.controlled_data --help
python -m trajnetdataset.convertThe above command performs the following operations:
- Step 1. readers.py: reads the raw data files and converts them to trackrows in .ndjson format
- Step 2. scene.py: prepares different scenes given the obtained trackrows
- Step 3. get_type.py: categorizes each scene based on our defined trajectory categorization
# create plots to check new dataset
python -m trajnetplusplustools.summarize output/train/*.ndjson
# obtain new dataset statistics
python -m trajnetplusplustools.dataset_stats output/train/*.ndjson
# visualize sample scenes
python -m trajnetplusplustools.trajectories output/train/*.ndjson- partial tracks are now included (for correct occupancy maps)
- pedestrians that appear in multiple chunks had the same id before (might be a problem for some input readers)
- explicit index of scenes with annotation of the primary pedestrian
# * the primary pedestrian has to move by more than 1 meter * at one point, the primary pedestrian has to be <3m away from another pedestrian
If you find this code useful in your research then please cite
@inproceedings{Kothari2020HumanTF,
title={Human Trajectory Forecasting in Crowds: A Deep Learning Perspective},
author={Parth Kothari and Sven Kreiss and Alexandre Alahi},
year={2020}
}
eth:
@article{Pellegrini2009YoullNW,
title={You'll never walk alone: Modeling social behavior for multi-target tracking},
author={Stefano Pellegrini and Andreas Ess and Konrad Schindler and Luc Van Gool},
journal={2009 IEEE 12th International Conference on Computer Vision},
year={2009},
pages={261-268}
}
ucy:
@article{Lerner2007CrowdsBE,
title={Crowds by Example},
author={Alon Lerner and Yiorgos Chrysanthou and Dani Lischinski},
journal={Comput. Graph. Forum},
year={2007},
volume={26},
pages={655-664}
}
wildtrack:
@inproceedings{chavdarova-et-al-2018,
author = "Chavdarova, T. and Baqué, P. and Bouquet, S. and Maksai, A. and Jose, C. and Bagautdinov, T. and Lettry, L. and Fua, P. and Van Gool, L. and Fleuret, F.",
title = {{WILDTRACK}: A Multi-camera {HD} Dataset for Dense Unscripted Pedestrian Detection},
journal = "Proceedings of the IEEE international conference on Computer Vision and Pattern Recognition (CVPR)",
year = 2018,
}
L-CAS:
@article{Sun20173DOFPT,
title={3DOF Pedestrian Trajectory Prediction Learned from Long-Term Autonomous Mobile Robot Deployment Data},
author={Li Sun and Zhi Yan and Sergi Molina Mellado and Marc Hanheide and Tom Duckett},
journal={2018 IEEE International Conference on Robotics and Automation (ICRA)},
year={2017},
pages={1-7}
}
CFF:
@article{Alahi2014SociallyAwareLC,
title={Socially-Aware Large-Scale Crowd Forecasting},
author={Alexandre Alahi and Vignesh Ramanathan and Fei-Fei Li},
journal={2014 IEEE Conference on Computer Vision and Pattern Recognition},
year={2014},
pages={2211-2218}
}
syi: Shuai Yi, Hongsheng Li, and Xiaogang Wang. Understanding Pedestrian Behaviors from Stationary Crowd Groups. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2015).edinburgh: B. Majecka, "Statistical models of pedestrian behaviour in the Forum", MSc Dissertation, School of Informatics, University of Edinburgh, 2009.