Research map / Image generation / Image restoration
Dual Residual Networks Leveraging the Potential of Paired Operations for Image Restoration
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace paired residual connections with the proposed dual residual connection (ablation) | DuRN task networks · Gaussian noise, real noise, motion blur, haze, raindrop, and rain-streak benchmarks | PSNR: 24.92 -> 25.05; 36.76 -> 36.84; 29.46 -> 29.90; 31.20 -> 32.60; 24.70 -> 25.32; 32.85 -> 33.21 | Helped |
| replace dual residual connection with paired-operation unit connection (b) (ablation) | DuRN task networks · Gaussian noise removal | PSNR: 25.05 -> 24.92 | Hurt |
| replace dual residual connection with paired-operation unit connection (b) (ablation) | DuRN task networks · Real noise removal | PSNR: 36.84 -> 36.76 | Hurt |
| replace dual residual connection with paired-operation unit connection (b) (ablation) | DuRN task networks · Motion blur removal | PSNR: 29.90 -> 29.46 | Hurt |
| replace dual residual connection with paired-operation unit connection (b) (ablation) | DuRN task networks · Haze removal | PSNR: 32.60 -> 31.20 | Hurt |
| replace dual residual connection with paired-operation unit connection (b) (ablation) | DuRN task networks · Raindrop removal | PSNR: 25.32 -> 24.70 | Hurt |
| replace dual residual connection with paired-operation unit connection (b) (ablation) | DuRN task networks · Rain-streak removal | PSNR: 33.21 -> 32.85 | Hurt |
| use up- and down-sampling paired operations rather than large/small convolutions (ablation) | DuRN motion-blur network · GoPro Dataset | PSNR: 29.40 -> 29.90 | Helped |
| add channel-wise attention to up/down-sampling blocks (ablation) | DuRN motion-blur network · GoPro Dataset | PSNR: 29.90 -> 29.96 | No clear effect |
| use up/down-sampling plus channel-wise attention rather than convolution pairs (ablation) | DuRN haze network · Dehaze Dataset | PSNR: 29.33 -> 32.60 | Helped |
| use channel-wise attention blocks rather than convolution-pair blocks (ablation) | DuRN raindrop network · RainDrop Dataset | PSNR: 24.69 -> 25.13 | Helped |
| use the task-designed DuRB-S instead of DuRB-P (ablation) | DuRN rain-streak network · DID-MDN Data | PSNR: 32.88 -> 33.21 | Helped |
Closest papers by meaning
- Image Restoration Using Very Deep Convolutional Encoder-Decoder Networks with Symmetric Skip Connections (NeurIPS 2016 · similarity 0.89)
- Residual-Conditioned Optimal Transport: Towards Structure-Preserving Unpaired and Paired Image Restoration (ICML 2024 · similarity 0.88)
- Attention-Based Adaptive Selection of Operations for Image Restoration in the Presence of Unknown Combined Distortions (CVPR 2019 · similarity 0.88)
- Omni-Kernel Network for Image Restoration (AAAI 2024 · similarity 0.88)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.87)
- Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.