jesssssse-z/ltcf

Long-Term Crowd Flow Dataset

This repository contains a dataset of 87,430 simulated crowd flows used in Laying the Foundations of Deep Long-Term Crowd Flow Prediction by Samuel S. Sohn, Honglu Zhou, Seonghyeon Moon, Sejong Yoon, Vladimir Pavlovic, and Mubbasir Kapadia. If you find this dataset useful in your research, please consider citing:

@inproceedings{sohnECCV20crowdflow,
  author    = {Sohn, Samuel S. and Zhou, Honglu and Moon, Seonghyeon and Yoon, Sejong and Pavlovic, Vladimir and Kapadia, Mubbasir},
  title     = {Laying the Foundations of Deep Long-Term Crowd Flow Prediction},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2020}
}

Installation

1. Clone this repository.

   >> git clone https://github.com/SSSohn/LTCF.git
   >> cd LTCF/
   

2. Download the pre-trained models from Google Drive and place them in the Models/Pre-Trained Models/ directory.

3. We recommend installing Python v3.6.5 from Anaconda and installing PyTorch (>= 1.1.0) by following the guidelines on the official instructions according to your specific CUDA version.

Training

If you want to reproduce the results of our pretrained models, first move the training and testing data into Models/Data/ and then run the following commands:

>> cd Models/
>> python Models/1_RunDataPreparation.py
>> python Models/2_RunTraining.py

This will train a new model for 200 epochs with a batch size of 32, learning rate of 0.01, and a momentum of 0.9.

Testing

In order to test the pre-trained models located in Models/Pre-Trained Models/ on one of the Testing sets, move the contents of the Testing set into Models/Data/Testing/ and run Models/3_RunTesting.py. This will convert the image files into Numpy files and predict the output using the specified model. The output images are saved in Models/Output/.

Folder Structure

+---Proxy Crowd Flow
|   +---CR 1.00
|   |   +---Testing
|   |   \---Training
|   +---CR 1.25
|   |   +---Testing
|   |   \---Training
|   +---CR 1.50
|   |   +---Testing
|   |   \---Training
|   +---CR 1.75
|   |   +---Testing
|   |   \---Training
|   \---CR 2.00
|       +---Testing
|       \---Training
\---Simulated Crowd Flow
    +---Multi-Goal Non-Uniform Agents
    |   +---Testing
    |   \---Training
    +---Non-Axis-Aligned Real Floorplans
    |   \---Testing
    \---Single-Goal Uniform Agents
        +---Testing
        \---Training

In each of the Training and Testing folders, there are 10 sub-folders, each corresponding to a component in the Framework image below:

+---A
+---A'
+---Cx'
+---Cy'
+---E
+---E'
+---G
+---G'
+---Y  (Ground Truth)
\---Y' (Compressed Ground Truth)

Dataset Sample

A	G	E	Cx'	Cy'	A'	G'	E'	Y	Y'

Framework

Acknowledgement

Part of our code is borrowed from the following repository:

• segnet_pytorch