andy-zzf/piepredict

PIE: A Large-Scale Dataset and Models for Pedestrian Intention Estimation and Trajectory Prediction

Pedestrian intention and trajectory estimation

This repository contains Python code and pretrained models for pedestrian intention and trajectory estimation presented in our paper A. Rasouli, I. Kotseruba, T. Kunic, and J. Tsotsos, "PIE: A Large-Scale Dataset and Models for Pedestrian Intention Estimation and Trajectory Prediction", ICCV 2019.

Table of contents

• Dependencies
• PIE dataset
• Train
• Test
• Citation
• Corresponding authors
• License

Dependencies

The interface is written and tested on Ubuntu 16.04 with Python 3.5, CUDA 9 and cuDNN 7. The interface also requires the following external libraries:

• tensorflow (tested with 1.9 and 1.14)
• keras (tested with 2.1 and 2.2)
• scikit-learn
• numpy
• pillow

To install via virtual environment (recommended) follow these steps:

• Install virtual environment sudo apt-get install virtualenv.

• Create a virtual environment with Python3:

> virtualenv --system-site-packages -p python3 ./venv
> source venv/bin/activate

• Install dependencies: pip3 install -r requirements.txt

PIE Dataset

The code is trained and tested with Pedestrian Intention Estimation (PIE) dataset.

Download annotations and video clips from the PIE webpage and place them in the PIE_dataset directory. The folder structure should look like this:

PIE_dataset
    annotations
        set01
        set02
        ...
    PIE_clips
        set01
        set02
        ...

Videos will be automatically split into individual frames for training. This will require 1.1T of free space on the hard drive.

Create environment variables for PIE data root and add them to your .bashrc:

export PIE_PATH=/path/to/PIE/data/root
export PIE_RAW_PATH=/path/to/PIE/data/PIE_clips/

Download PIE data interface pie_data.py from PIE github.

Train

To train all models from scratch and evaluate them on the test data use this command:

python train_test.py 1

This will train intention, speed and trajectory models separately and evaluate them on the test data.

_Note: training intention model uses image data and requires 32GB RAM.

Due to the random initialization of the networks and minor changes to the annotations there might be slight variation in the results.

Test

To reproduce the results of our best model which combines pedestrian intention and vehicle speed for pedestrian trajectory prediction run this command:

python train_test.py 2

Citation

If you use our dataset and/or models, please cite:

@inproceedings{rasouli2017they,
  title={PIE: A Large-Scale Dataset and Models for Pedestrian Intention Estimation and Trajectory Prediction},
  author={Rasouli, Amir and Kotseruba, Iuliia and Kunic, Toni and Tsotsos, John K},
  booktitle={ICCV},
  year={2019}
}

Corresponding authors

• Amir Rasouli
• Iuliia Kotseruba

Should you have any questions about downloading and using the data or the models, please feel free to raise an issue or send an email to [email protected] or [email protected].

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details