DGFNet: Multi-Agent Trajectory Prediction with Difficulty-Guided Feature Enhancement Network

Paper | Webpage

This is the official implementation of the paper Multi-Agent Trajectory Prediction with Difficulty-Guided Feature Enhancement Network.

This paper is currently being submitted to IEEE Robotics and Automation Letters (RA-L).

Introduction

Traditional methods usually perform holistic inference on the trajectories of agents, neglecting differences in prediction difficulty among agents. This paper proposes a novel Difficulty-Guided Feature Enhancement Network (DGFNet), which leverages the prediction difficulty differences among agents for multiagent trajectory prediction.

Firstly, we employ Spatio-temporal Feature Extraction to capture rich spatio-temporal features. Secondly, a Difficulty-Guided Decoder controls the flow of future trajectories into subsequent modules, obtaining reliable future trajectories. Then, feature interaction and fusion are performed through the Future Feature Interaction module. Finally, the fused actor features are fed into the Final Decoder to generate the predicted trajectory distributions for multiple participants.

Compared to the SOTA methods, our method balances trajectory prediction accuracy and real-time inference speed.

Argoverse 1(Single model)

• Performance Metrics:

Split	brier-minFDE	minFDE	MR	minADE	Param
Val	1.499	0.897	0.073	0.634	4.53
Test	1.742	1.117	0.108	0.763	-

Argoverse 1(Ensemble model--Five models trained from different random seeds)

• Performance Metrics:

Split	brier-minFDE	minFDE	MR	minADE
Test	1.693	1.110	0.107	0.752

Qualitative Results

• On Argoverse 1 motion forecasting dataset

• On Argoverse 2 motion forecasting dataset

---

Gettting Started

Install dependencies

• Create a new conda virtual env

conda create --name DGFNet python=3.8
conda activate DGFNet

• Install PyTorch according to your CUDA version. We recommend CUDA >= 11.1, PyTorch >= 1.8.0.

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 scikit-image IPython tqdm ipdb tensorboard

Train from scratch

• Preprocess full Argoverse 1 motion forecasting dataset using the script:

sh scripts/argo_preproc.sh

• Launch training using the script:

sh scripts/DGFNet_train.sh

• For model evaluation, please refer to the following scripts:

sh scripts/DGFNet_eval.sh

Test from scratch

• Generate files that can be submitted on the EvalAI:

sh scripts/DGFNet_test.sh

Contact

If you have any questions, please contact Guipeng Xin via email ([email protected]).

Citation

If you find DGFNet is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

@article{xin2024multi,
  title={Multi-Agent Trajectory Prediction with Difficulty-Guided Feature Enhancement Network},
  author={Xin, Guipeng and Chu, Duanfeng and Lu, Liping and Deng, Zejian and Lu, Yuang and Wu, Xigang},
  journal={arXiv preprint arXiv:2407.18551},
  year={2024}}

Acknowledgment

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

• Simpl
• ADAPT
• argoverse

License

This repository is licensed under MIT license.