AVDNet

[IEEE Signal Processing Letters, 2025] Adaptive Video Demoiréing Network with Subtraction-Guided Alignment
Seung-Hun Ok, Young-Min Choi, Seung-Wook Kim, Se-Ho Lee
Paper | Supplementary Materials

Introduction

<p align="center"> <img src="network.png" alt="AVDNet Architecture"/> </p> We propose an adaptive video demoiréing network (AVDNet) that effectively suppresses moiré artifacts while preserving temporal consistency. AVDNet transforms moiré-contaminated frames into temporally consistent clean frames by employing two key components: the adaptive bandpass block (ABB) and the subtraction-guided alignment block (SGAB). First, ABB applies an adaptive bandpass filter (ABF) to each frame, modulated by input-specific coefficients to selectively attenuate moiré frequencies based on the spectral distribution of the input. Then, SGAB aligns consecutive frames by exploiting subtraction maps, which effectively suppresses the propagation of moiré artifacts across time. Experimental results show that AVDNet outperforms existing video demoiréing methods, while maintaining a compact and efficient network architecture.

Environment

The experiments were conducted using the following software environment and libraries:

• Python: 3.11.4
• CUDA: 12.1
• PyTorch: 2.1.0
• Torchvision: 0.16.0
• numpy: 1.26.4
• scikit-image
• opencv-python
• deepspeed
• lpips
• tensorboard
• wandb

Dataset

Our project is based on the VDmoire dataset, which can be downloaded from here.
After downloading the dataset, place the folders as follows:

project_root/
    ├── AVDNet/
    │   ├── experiments/...
    │   │   ...
    │   └── train.py
    └── datasets/
        ├── homo/...
        └── optical/
            ├── iphone/...
            └── tcl/...

Pretrained Models

You can download the pretrained model for testing from here.
After downloading the model, place it in the AVDNet/experiments/ directory before running the test.

Note: If you use the pretrained model provided, be sure to set strict_load: false in the test option file, as some class names differ slightly.

Train/Test

The following is an example command for training on the iPhone-V1 subset using GPU 0:

CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/Train_ipv1.yml

The following is an example command for testing on the TCL-V2 subset using GPU 3:

CUDA_VISIBLE_DEVICES=3 python test.py -opt options/test/Test_tclv2.yml

You can run training or testing by selecting the appropriate .yml configuration file and specifying the GPU to use.

Results

<p align="center"> <img src="results.png" alt="Results"/> </p>

Citation

Please cite the following paper if you use this code in your research:

@article{ok2025adaptive,
title = {Adaptive Video Demoiréing Network With Subtraction-Guided Alignment},
author = {Ok, Seung-Hun and Choi, Young-Min and Kim, Seung-Wook and Lee, Se-Ho},
journal = {IEEE Signal Processing Letters},
volume = {32},
pages = {2733--2737},
year = {2025}
}

Acknowledgement

Our work and implementation were inspired by MBCNN and DTNet.
We sincerely thank the authors for making their code publicly available.

Contact

For any questions, please contact: cornking123@jbnu.ac.kr