Research map / Image generation / Image restoration
Robust Reflection Removal With Flash-Only Cues in the Wild
Chenyang Lei, Xudong Jiang, Qifeng Chen
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 |
|---|---|---|---|
| use flash-only image instead of ambient image alone as network input (ablation) | g_R+g_T · FOR-Aligned | PSNR: 25.13 -> 29.76 | Improved |
| use flash-only image instead of flash image as input (ablation) | g_R+g_T · FOR-Aligned | PSNR: 27.21 -> 29.76 | Improved |
| use flash-only image instead of flash image as input (ablation) | g_B naive model · FOR-Aligned | PSNR: 26.99 -> 27.55 | Improved |
| use reflection-pass two-stage network instead of single-network naive model (ablation) | g_B with flash-only input · FOR-Aligned | PSNR: 27.55 -> 29.76 | Improved |
| use reflection-pass two-stage network instead of single-network naive model (ablation) | g_B with flash-only input · FOR-Aligned | LPIPS: 0.187 -> 0.156 | Improved |
| add depth-based misaligned training data (ablation) | Model without alignment · FOR-Misaligned | PSNR: 25.41 -> 27.04 | Improved |
| use depth-based instead of homography-based misaligned data synthesis (ablation) | Model with alignment · FOR-Misaligned | PSNR: 27.58 -> 28.60 | Improved |
| add alignment module (finetuned optical flow) (ablation) | Model trained on misaligned data · FOR-Misaligned | PSNR: 27.04 -> 28.60 | Improved |
| use optical flow instead of homography for alignment (ablation) | Alignment module · FOR-Misaligned | PSNR: 27.97 -> 28.19 | Improved |
| finetune PWC-Net on reflection dataset (ablation) | Optical flow alignment · FOR-Misaligned | PSNR: 28.19 -> 28.60 | Improved |
| add test-time alignment only (ablation) | Model trained on aligned data · FOR-Misaligned | PSNR: 25.41 -> 26.50 | Improved |
| compute flash-only image instead of using flash image for linear RGB input (ablation) | Linear RGB model · Linear RGB images | PSNR: 30.88 -> 30.86 | No clear effect |
Closest papers by meaning
- Robust Reflection Removal With Reflection-Free Flash-Only Cues (CVPR 2021 · similarity 0.98)
- Flash-Split: 2D Reflection Removal with Flash Cues and Latent Diffusion Separation (CVPR 2025 · similarity 0.91)
- Revisiting Single Image Reflection Removal In the Wild (CVPR 2024 · similarity 0.88)
- PolarFree: Polarization-based Reflection-Free Imaging (CVPR 2025 · similarity 0.88)
- Removing Reflections from RAW Photos (CVPR 2025 · similarity 0.87)
- FIRM: Flexible Interactive Reflection ReMoval (AAAI 2025 · similarity 0.87)
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.
