Recurrent Flow Networks

This repository is the official implementation of the RFN, from Recurrent Flow Networks: A Latent Variable Model for Spatio-Temporal Density Modelling.

Full paper is available here

Real Data	Samples from model

\

Summary

This repository contains:

1. model.py: RFN model code
2. util.py: utility code
3. rfn_saved, transforms_saved, bns_saved: pre-trained version of the modules characterizing the RFN
4. /data: folder containing data used for the NYC-P experiment

Training and Evaluation code

A working Jupyter Notebook is provided in rfn_nyp.ipynb, showing a basic usage of the proposed RFN for the NYC-P task (more details in Section 3 of the paper).

The notebook contains:

1. Loading and processing of data
2. Building RFN object
3. Training/Loading pre-trained model code
4. Evaluation code
5. Basic visualizations