Image DeMoiréing Using Dual Camera Fusion on Mobile Phones

IEEE International Conference on Multimedia & Expo 2025 (ICME 2025)

Dataset | Paper

Overview

We propose an efffcient DCID method, where a lightweight UW image encoder is integrated into an existing deMoiréing network and a fast two-stage image alignment manner is present. Moreover, we construct a large-scale real-world dataset with diverse mobile phones and monitors, containing about 9,000 samples. We hope our investigation could inspire more future research in this more practical setting in image demoiréing.

Datasets

Download Links:

1. Baidu disk: here (Code: 9aab )

2. HuggingFace: here

Train

The configuration file (config) uses the parameter TRAIN_WORD to denote three different datasets, with each value corresponding to a specific dataset as follows:

• xiaomi: Represents the xiaomi dataset.
• iphone: Represents the iphone dataset.
• p40: Represents the huawei dataset.

The parameter align determines whether to use pre-generated alignment results from Keypoint Matching based Alignment(KMA) during training:

• False: The network will perform the first-stage alignment internally during the training process.
• True: The network will directly use the precomputed alignment results from the align folder. Specifically, it will use the file named w_align.jpg in the align directory for alignment, skipping the KMA during training.

python train.py --config ./config/dcid_config.yaml

Test

pre-trained models on our three datasets: here (Code: 9mqy )

python test.py --config ./config/test_dcid_config.yaml

Acknowledgement

This project is built based on UHDM.

Citations

@misc{mei2025imagedemoireingusingdual,
      title={Image Demoir\'eing Using Dual Camera Fusion on Mobile Phones}, 
      author={Yanting Mei and Zhilu Zhang and Xiaohe Wu and Wangmeng Zuo},
      year={2025},
      eprint={2506.08361},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2506.08361}, 
}