amandajshao / www_deep_crowd

The source code for our CVPR 2015 work "Deeply Learned Attributes for Crowded Scene Understanding" with a two-branch CNN model.

Deeply Learned Attributes for Crowded Scene Understanding

This is the source code for "Deeply Learned Attributes for Crowded Scene Understanding".

Features

Two-branch CNN model (i.e. appearance branch and motion branch)

Multi-class (i.e. 94 crowd attributes)

Files

Caffe Model

Three models: 
	single-branch (appearance model) `data_rgb_all_www_model_upgrade.caffemodel`
	single-branch (motion model) `data_motion_all_www_model_upgrade.caffemodel`
	two-branch (fusing appearance and motion models) `data_rgbm_all_www_top_combine_model_upgrade.caffemodel`

Prototxt

Motion channels

Training Data Splits

Project Site

Citation

J. Shao, K. Kang, C. C. Loy, and X. Wang Deeply Learned Attributes for Crowded Scene Understanding. Computer Vision and Pattern Recognition (CVPR), 2015.

@article{shao2015www,
  title={Deeply learned attributes for crowded scene understanding},
  author={Shao, Jing and Kang, Kai and Loy, Chen Change and Wang, Xiaogang},
  booktitle={Computer Vision and Pattern Recognition (CVPR)},
  year={2015}
}