/Scene-LSTM

Data and Code for "Scene-LSTM: A model for human trajectory prediction" (ISVC 2019)

Primary LanguagePython

Scene-LSTM

This code/implementation is available for research purposes. If you are using this code/data for your work, please cite the following paper:

Huynh, Manh, and Gita Alaghband. "Trajectory prediction by coupling scene-LSTM with human movement LSTM." International Symposium on Visual Computing. Springer, Cham, 2019.

The code are a bit messy for now, we are doing some clean-ups. If you have any questions/inquiries, please send email to manh.huynh@ucdenver.edu.

alt text

This repository contains

✔️ Processed data (in pixel and meter metrics) for ETH and UCY datasets. This data was also used in SGAN method.

├── data 
│     ├── pixel/*.txt
│     ├── meter/*.txt

✔️ Scripts to convert pixel to meter and vice versa. Double check the paths to homography matrices and input files.

├── data_utils
│     ├── homography_matrix/*.txt
│     ├── eth_utils/*.m (matlab scripts to process eth data)
│     ├── data_utils/*.m (matlab scripts to process ucy data))

All of the following items are available:

  • Visualization scripts.
  • My implementation for LSTM.
  • My implementation for Social-LSTM.
  • My implementation for Scene-LSTM.

Dependencies

The code was tested on Centos 7.0 with python 3.6.8 and pytorch 1.0.0.
The environment was setup using conda 4.5.12. Additional packages should be installed.