Research map / Image generation / Image restoration
Single Image Reflection Removal Through Cascaded Refinement
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 |
|---|---|---|---|
| Use cascaded refinement with both transmission and reflection sub-networks rather than removing G_R (ablation) | IBCLN · Nature, Zhang et al., and SIR² | PSNR: 21.79 -> 23.57; 20.65 -> 21.86; 22.36 -> 24.20 | Helped |
| Use iterative refinement rather than one time step (ablation) | IBCLN · Nature, Zhang et al., and SIR² | PSNR: 21.82 -> 23.57; 20.49 -> 21.86; 23.09 -> 24.20 | Helped |
| Add adversarial loss L_adv (ablation) | IBCLN · Nature and Zhang et al. | PSNR: 23.24 -> 23.57; 21.74 -> 21.86 | Helped |
| Add adversarial loss L_adv (ablation) | IBCLN · SIR² | SSIM: 0.885 -> 0.884 | Hurt |
| Add residual reconstruction loss L_residual (ablation) | IBCLN · Nature, Zhang et al., and SIR² | PSNR: 22.54 -> 23.57; 20.98 -> 21.86; 23.74 -> 24.20 | Helped |
| Add multi-scale perceptual loss L_MP (ablation) | IBCLN · Nature, Zhang et al., and SIR² | PSNR: 23.14 -> 23.57; 21.47 -> 21.86; 22.96 -> 24.20 | Helped |
| Combine all loss terms rather than use only pixel loss (ablation) | IBCLN · Nature | PSNR: 21.98 -> 23.57 | Helped |
| Set the total cascade steps to 3 rather than fewer or more steps (ablation) | IBCLN · SIR² | PSNR | Helped |
| Allow the network to learn the total number of steps automatically for each image (ablation) | IBCLN · Reflection removal images | performance | Hurt |
| Use IBCLN instead of ERRNet-F | Reflection removal · Average of five real-world test datasets | PSNR: 23.45 -> 24.08 | Helped |
| Use IBCLN instead of ERRNet-F | Reflection removal · Zhang et al. | PSNR: 23.35 -> 21.86 | Hurt |
Closest papers by meaning
- Single Image Reflection Removal Beyond Linearity (CVPR 2019 · similarity 0.90)
- Reversible Decoupling Network for Single Image Reflection Removal (CVPR 2025 · similarity 0.86)
- Revisiting Single Image Reflection Removal In the Wild (CVPR 2024 · similarity 0.85)
- FIRM: Flexible Interactive Reflection ReMoval (AAAI 2025 · similarity 0.85)
- Flash-Split: 2D Reflection Removal with Flash Cues and Latent Diffusion Separation (CVPR 2025 · similarity 0.84)
- Removing Reflections from RAW Photos (CVPR 2025 · similarity 0.84)
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.