Research map / Image generation / Image restoration
Self-Generated Defocus Blur Detection via Dual Adversarial Discriminators
Wenda Zhao, Cai Shang, Huchuan Lu
Check 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 a second discriminator to judge the blurred composite as well as the clear composite (ablation) | Self-generated DBD · CUHK and DUT | F-measure: CUHK: 0.360 -> 0.719 (single Dc); 0.353 -> 0.719 (single Db). DUT: 0.372 -> 0.683 (single Dc); 0.367 -> 0.683 (single Db | Improved |
| Add a second discriminator to judge the blurred composite as well as the clear composite (ablation) | Self-generated DBD · CUHK and DUT | MAE: CUHK: 0.264 -> 0.148 (single Dc); 0.699 -> 0.148 (single Db). DUT: 0.282 -> 0.190 (single Dc); 0.686 -> 0.190 (single Db | Improved |
| Add the bilateral triplet-excavating constraint with η=0.010 instead of η=0.000 (ablation) | Self-generated DBD · DUT | F-measure: 0.683 -> 0.701 | Improved |
| Add the bilateral triplet-excavating constraint with η=0.010 instead of η=0.000 (ablation) | Self-generated DBD · DUT | MAE: 0.190 -> 0.172 | Improved |
| Increase the bilateral triplet-excavating loss weight from η=0.010 to η=0.050 (ablation) | Self-generated DBD · CUHK and DUT | F-measure: CUHK: 0.769 -> 0.738; DUT: 0.701 -> 0.673 | Hurt |
| Increase the bilateral triplet-excavating loss weight from η=0.010 to η=0.050 (ablation) | Self-generated DBD · CUHK and DUT | MAE: CUHK: 0.119 -> 0.133; DUT: 0.172 -> 0.182 | Hurt |
| Use natural full-clear and full-blurred images in FCFB rather than simulated blur training data (ablation) | Self-generated DBD · CUHK and DUT | F-measure: FCSB(5D) to FCFB: CUHK 0.768 -> 0.769; DUT 0.655 -> 0.701 | Improved |
| Use natural full-clear and full-blurred images in FCFB rather than simulated blur training data (ablation) | Self-generated DBD · CUHK and DUT | MAE: FCSB(5D) to FCFB: CUHK 0.123 -> 0.119; DUT 0.192 -> 0.172 | Improved |
| Use more simulated blur degrees in FCSB, from FCSB(1D) to FCSB(5D) (ablation) | Self-generated DBD · CUHK and DUT | F-measure: CUHK: 0.715 -> 0.768; DUT: 0.610 -> 0.655 | Improved |
Closest papers by meaning
- United Defocus Blur Detection and Deblurring via Adversarial Promoting Learning (ECCV 2022 · similarity 0.92)
- Defocus Blur Detection via Multi-Stream Bottom-Top-Bottom Fully Convolutional Network (CVPR 2018 · similarity 0.91)
- Rethinking the Defocus Blur Detection Problem and A Real-Time Deep DBD Model (ECCV 2020 · similarity 0.91)
- Defocus Blur Detection via Multi-Stream Bottom-Top-Bottom Network (TPAMI 2019 · similarity 0.90)
- Prior and Prediction Inverse Kernel Transformer for Single Image Defocus Deblurring (AAAI 2024 · similarity 0.87)
- Multi-Focus Image Fusion via Explicit Defocus Blur Modelling (AAAI 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.
