Discrete-Constrained Regression for Local Counting Models (dcreg)

This repository contains the original pytorch code for our paper "Discrete-Constrained Regression for Local Counting Models" in ECCV 2022.

Prepare environment

conda env create -f requirements.yaml

Activate environment

conda activate dcreg

Download raw dataset

Raw JHU dataset could be obtained from the link.

preprocess dataset

Run jhu_main_final.m in Matlab.

Organize files

--> jhu_crowd_v2.0 (raw dataset)
-->data
   -->JHU_resize (processed dataset)
-->Models
   --> JHU
      --> best_epoch.pth

For training

sh train.sh

For testing

Download trained models from the link and put the file in "Models/JHU".

sh test.sh

You will get MAE 64.361 and MSE 281.078.

Sythesized Cell Dataset

A sample of synthesized dataset could be accessed from the link. More sythesized cell image could be generated with the code in the "generate_simulated_dataset.zip" file.

References

If you find this work or code useful for your research, please cite:

@inproceedings{xhp2022dcreg,
  title={Discrete-Constrained Regression for Local Counting Models},
  author={Xiong, Haipeng  and Yao, Angela},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2022},
  pages = {XXXX-XXXX}
}