🐉 DRG: Dynamic Relational Graph Modeling for Multi-Agent Trajectory Prediction

This is the official code implementation for manuscript "Dynamic Relational Graph Modeling for Multi-Agent Trajectory Prediction".

Visualization

Getting Started

Install dependencies

• Create a new conda virtual env

conda create --name DRG python=3.9
conda activate DRG

• Install PyTorch according to your CUDA version. We use CUDA = 12.4, PyTorch_Lighting = 2.4.0

conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.6 -c pytorch -c conda-forge

• Install Argoverse 1 APIs, please follow argoverse-api.

• Install other dependencies

  pip install -r requirements.txt 
  

Data preparation

• Refer to Simpl

Train from scratch

Once you prepare the dataset. Run train_ddp.py

Acknowledgment

We would like to express sincere thanks to the authors of the following packages and tools:

• Simpl
• HiVT
• LaneGCN

License

This repository is licensed under MIT license.