Hierarchical Motion Encoder-Decoder Network for Trajectory Forecasting (HMNet)

Our paper: https://arxiv.org/abs/2111.13324
We will release the complete version later

Setup：

The code was written in the following environment:

• python 3.7.11
• pytorch 1.10.0
• cuda 11.3
• cudnn 8.2.0

Preparation for data:

The raw data of Next Generation Simulation (NGSIM) is downloadable at https://ops.fhwa.dot.gov/trafficanalysistools/ngsim.htm

• Put the raw data into ./raw
• Run preprocess_data.m to preprocess the data for HMNet

Using the code:

To use the pretrained models at ./trained_models and evaluate the models performance run:

python evaluate.py