Research map / Architectures and efficiency / RNNs and normalization
A General Decoupled Learning Framework for Parameterized Image Operators
Qingnan Fan, Dongdong Chen, Lu Yuan, Gang Hua, Nenghai Yu, Baoquan 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 |
|---|---|---|---|
| train one network on random continuous parameter values instead of one network per value (L0) (ablation) | decoupled framework · L0 filter | PSNR avg: 36.11 -> 35.25 | Hurt |
| train one network on numerous parameter values (WLS) (ablation) | decoupled framework · WLS filter | SSIM avg: 0.990 -> 0.990 | No clear effect |
| joint training on 6/4 operators rather than single operator (ablation) | decoupled framework · 10 operators | PSNR avg: 34.40 -> 32.72 | Hurt |
| joint training on all 10 operators rather than single operator (ablation) | decoupled framework · 10 operators | PSNR avg: 34.40 -> 32.18 | Hurt |
| use decoupled framework instead of prior baseline for 10 jointly trained operators | joint 10 operators · 10 operators | PSNR avg: 31.02 -> 32.18 | Improved |
| use decoupled framework instead of DBL | image operator reproduction · 10 operators | PSNR avg: 26.92 -> 31.87 | Improved |
| place learned parameters deeper in network (norm layer 1 to 19) (ablation) | weight learning network · 10 operators | PSNR avg: 31.21 -> 28.67 | Hurt |
| learn conv weights rather than instance norm parameters at layer 19 (ablation) | weight learning network · 10 operators | PSNR avg: 28.67 -> 30.85 | Improved |
| learn norm in all layers rather than first layer only (ablation) | weight learning network · 10 operators | PSNR avg: 31.21 -> 31.62 | No clear effect |
| learn all conv layers instead of all norm layers (ablation) | weight learning network · 10 operators | PSNR avg: 31.62 -> 31.63 | No clear effect |
| add weight learning network to base network (single parameter) (ablation) | base network · WLS filter | PSNR avg: 40.54 -> 41.52 | Improved |
| add a second FC layer to weight learning network (ablation) | weight learning network · WLS filter | PSNR avg: 40.91 -> 39.67 | No clear effect |
Closest papers by meaning
- Decouple Learning for Parameterized Image Operators (ECCV 2018 · similarity 0.97)
- Decoupled Networks (CVPR 2018 · similarity 0.86)
- Learning to Learn Parameterized Classification Networks for Scalable Input Images (ECCV 2020 · similarity 0.86)
- Parameters as Experts: Adapting Vision Models with Dynamic Parameter Routing (ICML 2026 · similarity 0.85)
- Decoupled Dynamic Filter Networks (CVPR 2021 · similarity 0.85)
- Multilinear Operator Networks (ICLR 2024 · similarity 0.83)
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.
