/DACG

This is the implementation of our paper entitled: "Dynamic attention-based CVAE-GAN for pedestrian trajectory prediction"

Primary LanguagePython

DACG

This is the implementation of our paper entitled: "Dynamic attention-based CVAE-GAN for pedestrian trajectory prediction"
image text
Figure: Overview of the proposed DACG model. Herein, the whole model consists of a generator and a discriminator. The generator is made up of five modules, namely Feature Encoder, Social Aggregator, Mode Estimator, Goal Estimator, and Trajectory Decoder. Red lines indicate the processes that appear in the training phase only.

Contents

folder

  • results: storing the stepwise model training results
  • sgan: containing the DACG model
  • Trajectron: containing the dataset and dataloader

File

  • main.py: the script for training
  • argument_parser_neighbor.py: the hyperparameters applied in the proposed DACG model

Model training

Please run python main.py --dataset_name name to train a DACG model from scratch
(Please choose the "name" from {eth,hotel,univ,zara1,zara2})

Citation

If you find this repository useful, please cite:
@article{zhou2022dynamic,
title={Dynamic attention-based CVAE-GAN for pedestrian trajectory prediction},
author={Zhou, Zhou and Huang, Gang and Su, Zhaoxin and Li, Yongfu and Hua, Wei},
journal={IEEE Robotics and Automation Letters},
volume={8},
number={2},
pages={704--711},
year={2022},
publisher={IEEE}}