STDNet-Crowd-Counting

This is the official Tensorflow implementation of the paper: "Spatiotemporal Dilated Convolution with Uncertain Matching for Video-based Crowd Estimation (T-MM 2021)".

Paper Link

Citation

If you use this code for your research, please cite our paper. Thank you!

@article{9316927,
  author={Yu-Jen Ma, Hong-Han Shuai,and Wen-Huang Cheng},
  journal={IEEE Transactions on Multimedia}, 
  title={Spatiotemporal Dilated Convolution with Uncertain Matching for Video-based Crowd Estimation}, 
  year={2021},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TMM.2021.3050059}
}

Code

Install Dependencies

The code is used with Python 3.6, requiring the packages listed below.

tensoflow==1.14.0
opencv-python
pillow
scipy
numpy

The packages can be easily installed by pip install.

Train

Download the preprocessed UCSD Dataset and the initial weight for the VGG backbone. Google Drive Link
Unzip the downloaded file and modify the path to the same directory of this repository.
Run the python file (including the evaluation on the testing set).

python Train.py

The other details will be updated soon.

to be continued.

stk513486 / STDNet