AttConvLSTM

This work "Modeling Citywide Crowd Flows using Attentive Convolutional LSTM" has been published in ICDE 2021.

📄 Description

We present a novel deep model for this task, called "AttConvLSTM", which leverages a convolutional LSTM (ConvLSTM), Convolutional Neural Networks (CNNs) along with an attention mechanism, where ConvLSTM keeps spatial information as intact as possible during sequential analysis, and the attention mechanism can focus important crowd flow variations which cannot be identified by the recurrent module.

🔧 Dependencies

• Python == 3.5 (Recommend to use Anaconda or Miniconda)
• Tensorflow == 1.4.1
• NVIDIA GPU (NVIDIA GTX TITAN XP) + CUDA 10

Installation

1. Clone repo

   git clone https://github.com/BIT-MCS/AttConvLSTM.git
   cd AttConvLSTM

2. Install dependent packages

   pip install -r requirements.txt
   

⚡ Quick Inference

Get the usage information of the project

cd code/AttConvLSTM/experiment/att_conv_lstm2/
python train.py -h

Then the usage information will be shown as following

usage: train.py [-h] PATH SEQ_LENGTH DATA_WIDTH

positional arguments: PATH The directory of saving model SEQ_LENGTH Total sequence length DATA_WIDTH The resolution of data

optional arguments: -h, --help show this help message and exit

💻 Training

We provide complete training codes for AttConvLSTM.
You could adapt it to your own needs.

1. You can modify the config files AttConvLSTM/code/AttConvLSTM/experiment/att_conv_lstm2/conf.py For example, you can set the batch size and training mode by modifying these lines

   [4]     'BATCH' : 32,
   [32]    'SAVE_MODEL' : True,
   [33]    'LOAD_MODEL' : False,
   [34]    'IS_TEST' : False,
   

2. Training

   python train.py 
   

🏁 Testing

1. Before testing, you should modify the file AttConvLSTM/code/AttConvLSTM/experiment/att_conv_lstm2/conf.py as:

   [32]    'SAVE_MODEL' : False,
   [33]    'LOAD_MODEL' : True,
   [34]    'IS_TEST' : True,
   

2. Testing

   python train.py
   

📜 Acknowledgement

Corresponding author: Chi Harold Liu.

📧 Contact

If you have any question, please email [email protected].

Paper

If you are interested in our work, please cite our paper as

@INPROCEEDINGS{9458664,
  author={Liu, Chi Harold and Piao, Chengzhe and Ma, Xiaoxin and Yuan, Ye and Tang, Jian and Wang, Guoren and Leung, Kin K.},
  booktitle={2021 IEEE 37th International Conference on Data Engineering (ICDE)}, 
  title={Modeling Citywide Crowd Flows using Attentive Convolutional LSTM}, 
  year={2021},
  pages={217-228},
  doi={10.1109/ICDE51399.2021.00026}
}