Model Description
A machine learning model for crowd counting
- Model type: image-classifier
- License: mit
Crowd Counting Model
The aim is to build a model that can estimate the amount of people in a crowd from an image-
The model was built using CSRNet a crowd counting neural network designed by Yuhong Li, Xiaofan Zhang and Deming Chen (https://github.com/leeyeehoo/CSRNet-pytorch)
Model Sources
- Repository: https://github.com/leeyeehoo/CSRNet-pytorch
Uses
This model was created in the spirit of creating a model capable of counting the amount of people in a crowd using images.
Direct Use
model = CSRNet()
checkpoint = torch.load("weights.pth")
model.load_state_dict(checkpoint)
model.predict()
Bias, Risks, and Limitations
Although the model can be very accurate its not exact, it has a 2%-6% error in the prediction.
Training Details
Training Data
The model was trained using the ShanghaiTech Dataset, specifically the Shanghai B Dataset.
Training Procedure
The info on training procedure can be found in this repository https://github.com/leeyeehoo/CSRNet-pytorch
Evaluation and Results
The model reached a MAE of 10.6
Citation
Model creation and training
@inproceedings{li2018csrnet, title={CSRNet: Dilated convolutional neural networks for understanding the highly congested scenes}, author={Li, Yuhong and Zhang, Xiaofan and Chen, Deming}, booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition}, pages={1091--1100}, year={2018} }
Dataset
@inproceedings{zhang2016single, title={Single-image crowd counting via multi-column convolutional neural network}, author={Zhang, Yingying and Zhou, Desen and Chen, Siqin and Gao, Shenghua and Ma, Yi}, booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition}, pages={589--597}, year={2016} }