Related Experiment Video
Updated: Apr 13, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
1.2K
High-resolution image deraining via dual-branch features interaction and fusion.
Weilong Huang1, Jiaxue Mei1, Tao Yan1
1School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, 214122, Jiangsu, China.
Summary
This study introduces the Dual-branch Deraining Network (DDNet) for high-resolution image deraining. DDNet effectively removes rain streaks while preserving image details, overcoming limitations of existing methods.
Area of Science:
- Computer Vision
- Image Processing
- Artificial Intelligence
Background:
- Existing image deraining methods struggle with high-resolution images due to increased computational costs and difficulty in capturing complex rain streak patterns.
- Current models often lead to over-smoothing and loss of critical details in high-resolution rainy images.
- There's a need for efficient and effective deraining solutions for high-resolution imagery.
Purpose of the Study:
- To propose an efficient Dual-branch Deraining Network (DDNet) for high-resolution image deraining.
- To enhance the representation capacity of Convolutional Neural Network (CNN) and Transformer branches through frequency-domain feature interaction and dynamic feature fusion.
- To address feature disparities between CNN and Transformer architectures.
Main Methods:
- Developed an efficient Dual-branch Deraining Network (DDNet) integrating CNN and Transformer branches.
- Introduced an Adaptive Frequency Reciprocal Module (AFRM) for adaptive frequency spectral filter generation and inter-branch interaction.
- Proposed a Discrete Cosine Transform (DCT) based Multi-Spectral Fusion Module (MSFM) for dynamic feature fusion.
- Implemented a Dual Differential Cross Attention Module (DDCAM) to mitigate feature discrepancies between branches.
Main Results:
- DDNet demonstrates effectiveness and efficiency in deraining high-resolution images.
- The proposed modules (AFRM, MSFM, DDCAM) contribute to improved feature representation and fusion.
- Experiments on real-world and synthetic datasets validate the model's performance.
Conclusions:
- DDNet offers a robust solution for high-resolution image deraining, preserving image quality and details.
- The integration of frequency-domain processing and cross-architecture attention mechanisms enhances deraining performance.
- The model provides an efficient alternative to existing methods for high-resolution image restoration.
