CAD_GD-CVPR25

(CVPR25) This repository is the official implementation of our Paper Exploring Contextual Attribute Density in Referring Expression Counting

Installation

Our code has been tested on Python 3.10 and PyTorch 2.4.0.

1. Install GroundingDINO.
2. Install requirements.txt.

Data Preparation

We train and evaluate our methods on REC8K and FSC-147 dataset. Please follow the REC8K and FSC-147 official repository to download and unzip the dataset.

• REC-8k
• FSC147
• FSC147-D

About the density map: Fot the FSC-147, we use the density map of FSC-147 directly. For the Rec8k, we generate the density map using fixed kernel size, you can download the generated density maps from the link.

Inference

You can run following command to conduct the inference on the REC-8k and FSC-147 dataset.

python test_rec.py
python test_fsc.py

Training

We use the pretrained model from GroundingDINO, please download the pretrained weight from GroundingDINO. Then you can run the following command to conduct the traininng on the REC-8k or FSC-147 dataset.

python train_rec8k.py
python train_fsc.py

Citation

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

@inproceedings{wang2025exploring,
  title={Exploring Contextual Attribute Density in Referring Expression Counting},
  author={Wang, Zhicheng and Pan, Zhiyu and Peng, Zhan and Cheng, Jian and Xiao, Liwen and Jiang, Wei and Cao, Zhiguo},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={19587--19596},
  year={2025}
}