Official PyTorch implementation of the paper "Efficient Masked AutoEncoder for Video Object Counting and A Large-Scale Benchmark".

百度网盘： Link 提取码：t3tv

Google Drive: Link

put the result .txt file on the root of dataset and run the script eval.py in toolkit folder.

create conda environment by

conda create --name <env_name> python=3.9
conda activate <env_name>
pip install -r requirements.txt

Follow the instructions to install pwcnet in folder './emac/models'

The folder structure should be like:

emac
└── models
    ├── correlation_package
    ├── __init__.py
    └── pwcnet.py

Download the pretrained weight of MultiMAE and PWC-Net to ./cfgs

bash density_opt.sh

Path of the config file should be set in density_opt.sh.

python test_opt.py

Some args in test_opt.py:

• 'image_height, image_width': the size of input images
• 'data_path': the path to the dataset
• 'dataset': the dataset name
• 'weight_path': the path to the model weights

@inproceedings{
cao2025efficient,
title={Efficient Masked AutoEncoder for Video Object Counting and A Large-Scale Benchmark},
author={Bing Cao and Quanhao Lu and Jiekang Feng and Qilong Wang and Pengfei Zhu and Qinghua Hu},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=sY3anJ8C68}
}

This code is based on the MultiMAE and PWC-Net. We thank the authors for their excellent work.

This work (dataset and code) is licensed under CC BY-NC-SA 4.0