PedstrianPrediction

This repo re-implement social LSTM and implement a new Spatial Pyramid Social LSTM proposed by us.
This implementation is based on https://github.com/vvanirudh/social-lstm-tf. We fix several bugs and vital errors in their repo. We also upgrade the tensorflow API to 1.8.0.

File Structure

data

• getPixelCoordinates.m: the matlab code to transform original ETH dataset to pixel_pos.csv, which is used in our code. This file is based on the referred implementation.
• pixel_pos.csv: the data file used by our code
• transformed_data.pkl: pixel_pos.csv will be transformed in our code and save as transformed_data.pkl

social_lstm

• DataLoader.py: deal with data loading and preprocess
• grid.py: calculate grid or pyramid mask, called by train.py
• model.py: IMPORTANT! all model (including social lstm and spatial pyramid social lstm) are defined here
• social_sample.py: predict/test code, could be called using proper console parameters (use social_sample.py --help to see)
• social_visualize.py: to draw predicted graphs
• train.py: train code, could be called using proper console parameters (use train.py --help to see)

plot

This directory contains several prediction plots for "Spatial Pyramid Social LSTM" method. Other method's plot can't be obtained since lab server is under maintenance (explained in our report).

save

This directory contains model file for "Spatial Pyramid Social LSTM" method. Other method's model can't be obtained since lab server is under maintenance (explained in our report).

Usage

1. delete everything under social_lstm/save/
2. run train.py to train a model
3. run social_sample.py to predict
4. run social_visualize.py to visualize

Authored by Letian Chen & Tianyang Zhao