Trajectory Prediction with Latent Space Energy-Based Model

Requirements:

torch==1.4.0
scipy==1.4.1
numpy==1.18.5
pandas==0.24.2

Environment

Code is tested on Ubuntu 16.04, Python 3.5, NVIDIA 1080 Ti, CUDA 10.1

Dataset

SDD is available in dataset/sdd.

Other datasets are in google drive: https://drive.google.com/drive/folders/1TaRvS_nA90KGKVgThIP-IIIbIKdmuIKe?usp=sharing

You can download and put them in dataset

Pretrained Models

A pretrained model for SDD is available in saved_models.

More models are in google drive: https://drive.google.com/drive/folders/1Pu7ggwIDknk5Xs-MfKQHNnH0-UdMa1E3?usp=sharing

You can download them and put them in saved_models.

Run Trained Models

# SDD
python lbebm_sdd.py

# ETH
python lbebm_eth.py

Train New Models

You can train new models by simply removing the model_path.

Change --dataset_name to train other datasets in eth-ucy.

# SDD
python lbebm_sdd.py --model_path ''

# ETH
python lbebm_eth.py --model_path ''

Contact

Please contact Bo Pang ([email protected]) for questions.