Research map / Image generation / Image restoration
Context-Aware Pretraining for Efficient Blind Image Decomposition
Chao Wang, Zhedong Zheng, Ruijie Quan, Yifan Sun, Yi Yang
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 the separation encoder EA (ablation) | CPNet · Task II, rainstreak + raindrop + snow | PSNR: 18.25 -> 28.57 | Improved |
| add the reconstruction encoder EB (ablation) | CPNet · Task II, rainstreak + raindrop + snow | PSNR: 21.36 -> 28.57 | Improved |
| add the feature interaction block (ablation) | CPNet · Task II, rainstreak + raindrop + snow | PSNR: 25.84 -> 28.57 | Improved |
| add the feature refinement block (ablation) | CPNet · Task II, rainstreak + raindrop + snow | PSNR: 27.23 -> 28.57 | Improved |
| add the feature enhancement block (ablation) | CPNet · Task II, rainstreak + raindrop + snow | PSNR: 27.47 -> 28.57 | Improved |
| add conditional attribute loss (ablation) | CPNet · Task II, rainstreak + raindrop + snow | PSNR: 28.29 -> 28.57 | Improved |
| pretrain on ImageNet instead of the BID dataset (ablation) | CPNet · Task I case (5) | PSNR: 26.80 -> 30.07 | Improved |
| pretrain on a broader set of ImageNet classes rather than only selected scene classes (ablation) | CPNet · Task I case (5) | PSNR: 28.43 -> 30.07 | Improved |
| pretrain on scene-relevant rather than selected irrelevant object classes (ablation) | CPNet · Task I case (5) | PSNR: 24.89 -> 28.43 | Improved |
| use 10% of ImageNet rather than the full ImageNet pretraining set (ablation) | CPNet · Task I case (5) | PSNR: 30.07 -> 28.95 | Hurt |
| use CPNet rather than BIDeN | blind image decomposition · CityScape Task I, case (6) | PSNR: 26.44 -> 29.57 | Improved |
| use CPNet rather than BIDeN for single-noise restoration after multi-noise training | BID models · Task II, raindrop | BRISQUE: 21.22 -> 20.16 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning a Discriminative Prior for Blind Image Deblurring (CVPR 2018 · similarity 0.87)
- Restore Anything with Masks: Leveraging Mask Image Modeling for Blind All-in-One Image Restoration (ECCV 2024 · similarity 0.87)
- Learning Spatially-Variant MAP Models for Non-Blind Image Deblurring (CVPR 2021 · similarity 0.87)
- DiffBIR: Toward Blind Image Restoration with Generative Diffusion Prior (ECCV 2024 · similarity 0.87)
- Cross-Scale Self-Supervised Blind Image Deblurring via Implicit Neural Representation (NeurIPS 2024 · similarity 0.87)
- MC-DiT: Contextual Enhancement via Clean-to-Clean Reconstruction for Masked Diffusion Models (NeurIPS 2024 · 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.
