Research map / Image generation / Image restoration
A Generic Deep Architecture for Single Image Reflection Removal and Image Smoothing
Qingnan Fan, Jiaolong Yang, Gang Hua, Baoquan Chen, David Wipf
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 predicted target edge from E-CNN (CEILNet vs I-CNN only) (ablation) | I-CNN · L0 smoothing | PSNR: 32.58 -> 37.10 | Improved |
| add predicted target edge from E-CNN (vs input edge, 64 layers) (ablation) | I-CNN 64 layers · L0 smoothing | PSNR: 34.86 -> 37.10 | Improved |
| use predicted target edge instead of input image edge (ablation) | I-CNN 64 layers · L0 smoothing | MSE: 22.50 -> 13.34 | Improved |
| stack more layers in I-CNN instead of E-CNN (64 layers vs 32 layers) (ablation) | I-CNN only · L0 smoothing | PSNR: 32.58 -> 33.33 | Improved |
| use 32-layer E-CNN plus 32-layer I-CNN instead of best 70-layer I-CNN (ablation) | CEILNet · L0 smoothing | PSNR: 33.37 -> 37.10 | Improved |
| replace Domain Transform with I-CNN given predicted edge (ablation) | CEILNet · L0 smoothing | PSNR: 31.17 -> 37.10 | Improved |
| use predicted edge instead of input image edge with Domain Transform (ablation) | Domain Transform · L0 smoothing | PSNR: 27.38 -> 31.17 | Improved |
| use CEILNet instead of Li and Brown for reflection removal | CEILNet · Reflection removal (100 synthetic images) | PSNR: 15.50 -> 18.55 | Improved |
| use CEILNet instead of Xu et al. for L0 smoothing | CEILNet · Image smoothing | PSNR: 31.66 -> 37.10 | Improved |
| use CEILNet instead of Liu et al. across four filters | CEILNet · Image smoothing vs Liu et al. [24] | PSNR: 35.64 -> 39.36 | Improved |
| use CEILNet instead of Liu et al. on RGF | CEILNet · Image smoothing vs Liu et al. [24] | SSIM: 0.986 -> 0.989 | Improved |
| use CEILNet instead of Xu et al. for running time at 720p | CEILNet · Running time 720p | seconds: 0.33 -> 0.35 | No clear effect |
Closest papers by meaning
- Seeing Deeply and Bidirectionally: A Deep Learning Approach for Single Image Reflection Removal (ECCV 2018 · similarity 0.92)
- Single Image Reflection Separation With Perceptual Losses (CVPR 2018 · similarity 0.90)
- Single Image Reflection Removal Exploiting Misaligned Training Data and Network Enhancements (CVPR 2019 · similarity 0.90)
- Single Image Reflection Removal Through Cascaded Refinement (CVPR 2020 · similarity 0.89)
- V-DESIRR: Very Fast Deep Embedded Single Image Reflection Removal (ICCV 2021 · similarity 0.89)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (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.
