Research map / Image generation / Image restoration
Single Image Reflection Removal Exploiting Misaligned Training Data and Network Enhancements
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 |
|---|---|---|---|
| add channel-wise context and multi-scale spatial context (ablation) | BaseNet on Real20 · Real20 | PSNR: 21.51 -> 22.89 | Helped |
| add channel-wise context (ablation) | BaseNet on Real20 · Real20 | PSNR: 21.51 -> 22.61 | Helped |
| add multi-scale spatial context (ablation) | BaseNet on Real20 · Real20 | PSNR: 21.51 -> 21.75 | Helped |
| add channel-wise and multi-scale spatial context together (ablation) | BaseNet on Synthetic · Synthetic | PSNR: 25.71 -> 27.88 | Helped |
| add 50 aligned real images (ablation) | ERRNet · Synthetic test images | PSNR: 19.79 -> 22.00 | Helped |
| increase aligned real training images from 50 to 90 (ablation) | ERRNet · Synthetic test images | PSNR: 22.00 -> 22.89 | Helped |
| use pixel-wise loss for 40 unaligned images instead of aligned-invariant feature loss (ablation) | ERRNet · Synthetic test images | PSNR: 22.00 -> 21.85 | Hurt |
| add 40 unaligned images supervised with alignment-invariant loss (ablation) | ERRNet · Synthetic test images | PSNR: 22.00 -> 22.38 | Helped |
| add 40 unaligned images supervised with contextual loss (ablation) | ERRNet · Synthetic test images | PSNR: 22.00 -> 22.47 | Helped |
| combine invariant and contextual losses on 40 unaligned images (ablation) | ERRNet · Synthetic test images | PSNR: 22.00 -> 22.43 | Helped |
| finetune using collected unaligned pairs and alignment-invariant loss | ERRNet · Collected unaligned image pairs | human preference score: 0.51 | Helped |
| finetune using collected unaligned pairs and alignment-invariant loss | BDN-F · Collected unaligned image pairs | human preference score: 0.62 | Helped |
Closest papers by meaning
- Single Image Reflection Separation With Perceptual Losses (CVPR 2018 · similarity 0.91)
- Single Image Reflection Removal Through Cascaded Refinement (CVPR 2020 · similarity 0.89)
- Revisiting Single Image Reflection Removal In the Wild (CVPR 2024 · similarity 0.88)
- Reversible Decoupling Network for Single Image Reflection Removal (CVPR 2025 · similarity 0.88)
- FIRM: Flexible Interactive Reflection ReMoval (AAAI 2025 · similarity 0.87)
- Single Image Reflection Separation via Dual-Stream Interactive Transformers (NeurIPS 2024 · similarity 0.85)
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.