Research map / Image generation / Image restoration
Learning to Jointly Generate and Separate Reflections
Daiqian Ma, Renjie Wan, Boxin Shi, Alex C. Kot, Ling-Yu Duan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add the edge map loss (ablation) | joint generate-and-separate model · SIR2 | PSNR: 23.52 -> 24.32 | Improved |
| add the edge map loss (ablation) | joint generate-and-separate model · SIR2 | SSIM: 0.873 -> 0.892 | Improved |
| add the edge map loss (ablation) | joint generate-and-separate model · SIR2 | SSIMr: 0.826 -> 0.858 | Improved |
| add the content reconstruction loss (ablation) | joint generate-and-separate model · SIR2 | SSIMr: 0.845 -> 0.858 | Improved |
| add the content reconstruction loss (ablation) | joint generate-and-separate model · SIR2 | SSIM: 0.886 -> 0.892 | Improved |
| add the content reconstruction loss (ablation) | joint generate-and-separate model · SIR2 | PSNR: 24.31 -> 24.32 | No clear effect |
| mix generated data (prob 0.7) with Eq. (1) synthesized data (prob 0.3) to train the separator | Ours · SIR2 | PSNR: 24.32 -> 24.48 | Improved |
| mix generated data with Eq. (1) data | Ours · SIR2 | SSIMr: 0.858 -> 0.870 | Improved |
| mix generated data with Eq. (1) data | Ours · SIR2 | SSIM: 0.892 -> 0.903 | Improved |
| train separator with generated reflections instead of only Eq. (1) synthesis | separator · SIR2 | PSNR: 24.06 -> 24.32 | Improved |
| use weakly supervised joint framework instead of Wan18 | reflection removal · SIR2 | PSNR: 24.08 -> 24.32 | Improved |
| use joint framework instead of unpaired CycleGAN mapping | reflection removal · SIR2 | PSNR: 20.10 -> 24.32 | Improved |
Closest papers by meaning
- Single Image Reflection Separation via Component Synergy (ICCV 2023 · similarity 0.90)
- Reflection Separation from a Single Image via Joint Latent Diffusion (CVPR 2026 · similarity 0.90)
- Single Image Reflection Removal Beyond Linearity (CVPR 2019 · similarity 0.90)
- Single Image Reflection Removal Through Cascaded Refinement (CVPR 2020 · similarity 0.89)
- Seeing Deeply and Bidirectionally: A Deep Learning Approach for Single Image Reflection Removal (ECCV 2018 · similarity 0.89)
- Revisiting Single Image Reflection Removal In the Wild (CVPR 2024 · 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.
