Awesome
Crowd-Counting-with-Deep-Negative-Correlation-Learning
The codes for CVPR-2018 paper "Crowd Counting with Deep Negative Correlation Learning" in http://openaccess.thecvf.com/content_cvpr_2018/papers/Shi_Crowd_Counting_With_CVPR_2018_paper.pdf
To run this codes, what you need to do is as follows:
-
Compiling the Caffe codes. You should have installed Caffe correctly. Then you can clone or download our codes and make some changes in Makefile.config to compile correctly.
-
Preparing your data. The codes for preparing crowd counting dataset can be found in ¨examples/crowd/shanghaiA/predata¨. If you would like to run your own tasks, you have to write the codes by yourself.
-
Training. In ¨examples/crowd¨, you can find the network prototxt and solver prototxt, and you should make some changes according to your tasks. If you want to use different ¨K¨ (K stands for the number of base regresors in the ensemble, we found K=64 gives the best performance in our tasks.), the ¨examples/crowd/shanghaiA/create_prototxt.py¨ can help to generate your network prototxt easily.
-
Testing. If you use MAE and MSE as your evaluation metrics, you can monitor the testing results in training. You just need to add ¨layer { name: "mae" type: "MAELoss" bottom: "avgscore" bottom: "label" top: "mae" include { phase: TEST } }¨ and ¨layer { name: "mse" type: "MSELoss" bottom: "avgscore" bottom: "label" top: "mse" include { phase: TEST } }¨ to your network prototxt.
-
Please carefully tune learning rate, we found it had a great influence to the training results in our tasks.
If you use this codes, please kindly cite our paper:
@inproceedings{Shi_2018_CVPR,
title={Crowd Counting With Deep Negative Correlation Learning},
author={Shi, Zenglin and Zhang, Le and Liu, Yun and Cao, Xiaofeng and Ye, Yangdong and Cheng, Ming-Ming and Zheng, Guoyan},
booktitle={CVPR},
year={2019}
}
@articles{zhang2019tpamicounting,
title={Nonlinear Regression via Deep Negative Correlation Learning},
author={Le Zhang, Zenglin Shi, Ming-Ming Cheng, Yun Liu, Jia-Wang Bian, Joey Tianyi Zhou, Guoyan Zheng, Zeng Zeng},
booktitle={TPAMI},
year={2019}
}
Please feel free to contact us if you still have any questions. Zenglin Shi: iezlshi@gmail.com Le Zhang: zhang.le@adsc.com.sg