Research map / Image generation / Image restoration
Seeing Deeply and Bidirectionally: A Deep Learning Approach for Single Image Reflection Removal
Jie Yang, Dong Gong, Lingqiao Liu, Qinfeng Shi
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 reflection estimator H to vanilla generator (ablation) | Vanilla G0 · Synthetic reflection test (500 images) | PSNR: 22.10 -> 22.30 | Improved |
| add background estimator G1 to form full bidirectional unit (ablation) | Vanilla G0 + H · Synthetic reflection test (500 images) | PSNR: 22.30 -> 23.11 | Improved |
| make the vanilla generator deeper instead of adding the bidirectional unit (ablation) | Vanilla G0 · Synthetic reflection test (500 images) | PSNR: 22.10 -> 22.16 | Improved |
| train jointly end-to-end instead of greedy module-wise training (ablation) | BDN · Synthetic reflection test (500 images) | PSNR: 20.82 -> 23.11 | Improved |
| train jointly from scratch instead of greedy training plus fine-tuning (ablation) | BDN · Synthetic reflection test (500 images) | PSNR: 22.43 -> 23.11 | Improved |
| add adversarial loss (ablation) | BDN · Synthetic reflection test (500 images) | PSNR: 23.06 -> 23.11 | No clear effect |
| add adversarial loss (ablation) | BDN · Synthetic reflection test (500 images) | SSIM: 0.833 -> 0.835 | No clear effect |
| use BDN instead of Fan et al. when trained on our synthetic data | BDN · Our synthetic dataset | PSNR: 20.03 -> 23.11 | Improved |
| use BDN instead of Fan et al. when trained on the dataset of Fan et al. | BDN · Synthetic dataset of Fan et al. | PSNR: 18.29 -> 20.82 | Improved |
| use BDN instead of Fan et al. on real SIR data | BDN · SIR dataset, Postcard | PSNR: 21.0829 -> 20.4076 | Hurt |
| use BDN instead of Fan et al. on real SIR data | BDN · SIR dataset, Solid objects | PSNR: 23.5324 -> 22.7076 | Hurt |
| use BDN instead of Fan et al. on real SIR data | BDN · SIR dataset, Wild scenes | PSNR: 22.0618 -> 22.1082 | Improved |
Closest papers by meaning
- A Generic Deep Architecture for Single Image Reflection Removal and Image Smoothing (ICCV 2017 · similarity 0.92)
- Single Image Reflection Removal Through Cascaded Refinement (CVPR 2020 · similarity 0.91)
- Single Image Reflection Removal Exploiting Misaligned Training Data and Network Enhancements (CVPR 2019 · similarity 0.90)
- CoRRN: Cooperative Reflection Removal Network (TPAMI 2019 · similarity 0.90)
- Single Image Reflection Separation With Perceptual Losses (CVPR 2018 · similarity 0.89)
- Revisiting Single Image Reflection Removal In the Wild (CVPR 2024 · 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.
