trajectory_prediction_TraPHic

This repository for studying [TraPHic paper (CVPR '18)] (https://arxiv.org/abs/1812.04767).
The script is originally forked from [this repository] (https://github.com/rohanchandra30/Spectral-Trajectory-and-Behavior-Prediction).
The script has been further revised in the network architecture, and it is still working on.

Depencencies:

• Python 3.5.3
• torch 1.3.1
• pytorch-ignite
• numpy 1.16.2

How to generate dataset:

In this script, it uses apolloscape [dataset] (http://apolloscape.auto/trajectory.html). The datasets can be preprocessed as people, vehicles and bicycles/motorcycles. It means if you want to train the network for predicting people only, change 'CLASS_TYPE' to 'human' in traphic_main.py. Change 'GENERATE_DATASET' as 'True', and run traphic_main.py.

How to train:

1. Clone this repository

$ git clone <repository>

2. Training

$ python3 traphic_main.py