Skip to content

Latest commit

 

History

History
56 lines (46 loc) · 1.78 KB

File metadata and controls

56 lines (46 loc) · 1.78 KB

lyft-motion-prediction-for-autonomous-vehicle

prediction

pred clip

Description

Build motion prediction models for self-driving vehicles to predict other car/cyclist/pedestrian (called "agent")'s motion.

The image from L5Kit official document: http://www.l5kit.org/README.html
Lyft official page: https://self-driving.lyft.com/level5/prediction/

Table of Contents

Environment Setup

  • Python 3.* is installed
  • Set permission
chmod 700 bin/bootstrap
  • Run the bootstrap for installing the requirements
bin/bootstrap

Dataset

Dataset is not present in this repo, please download the Lyft Level 5 Prediction Dataset kit from the official website, and cite the following in your work:

@misc{lyft2020,
  title = {One Thousand and One Hours: Self-driving Motion Prediction Dataset},
  author = {Houston, J. and Zuidhof, G. and Bergamini, L. and Ye, Y. and Jain, A. and Omari, S. and Iglovikov, V. and Ondruska, P.},
  year = {2020},
  howpublished = {\url{https://level5.lyft.com/dataset/}}
}

Unzip the the dataset into the folder dataset/.

Training

  • After installing all the requirements, run the following command for trainig
python train.py -d -gpu -model MODEL_NAME
  • -d: debug mode, default is False
  • -gpu: train on GPU, default is on CPU
  • model: REQURIED, all the available models are in the folder model/, simply input the name of the model file.
    (eg. -model baseline for model baseline.py)

Prediction