Research map / Image generation / Image restoration
Generating Aligned Pseudo-Supervision From Non-Aligned Data for Image Restoration in Under-Display Camera
Ruicheng Feng, Chongyi Li, Huaijin Chen, Shuai Li, Jinwei Gu, Chen Change Loy
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 |
|---|---|---|---|
| Generate aligned pseudo targets with AlignFormer rather than using real misaligned pairs and pixel loss | PPM-UNet training · UDC real test set | PSNR: 19.04 -> 22.95 | Improved |
| Train on real captured data rather than the synthetic dataset | PPM-UNet · UDC real test set | PSNR: 19.03 -> 19.04 | Improved |
| Use contextual loss on non-aligned real pairs instead of pixel reconstruction loss | PPM-UNet training · UDC real test set | PSNR: 19.04 -> 20.85 | Improved |
| Use CoBi loss on non-aligned real pairs instead of pixel reconstruction loss | PPM-UNet training · UDC real test set | PSNR: 19.04 -> 21.57 | Improved |
| Train DISCNet on AlignFormer-generated pseudo pairs | DISCNet · UDC real test set | PSNR: 19.25 -> 21.66 | Improved |
| Add RAFT flow guidance instead of zero flow (ablation) | AlignFormer · UDC-reference alignment | PCK (α = 0.01): 9.61 -> 58.75 | Improved |
| Use RAFT rather than SPyNet flow guidance (ablation) | AlignFormer · UDC-reference alignment | PCK (α = 0.01): 31.31 -> 58.75 | Improved |
| Use RAFT rather than PWC-Net flow guidance (ablation) | AlignFormer · UDC-reference alignment | PCK (α = 0.01): 33.66 -> 58.75 | Improved |
| Add the Domain Alignment Module (ablation) | AlignFormer · UDC-reference alignment | PCK (α = 0.01): 56.15 -> 58.75 | Improved |
| Use local-grid radius 2 rather than radius 0 (ablation) | AlignFormer · UDC-reference alignment | PCK (α = 0.10): 98.71 -> 99.93 | Improved |
| Use local-grid radius 2 rather than radius 1 (ablation) | AlignFormer · UDC-reference alignment | PCK (α = 0.03): 95.07 -> 95.08 | No clear effect |
| Add the Pyramid Pooling Module (ablation) | PPM-UNet · UDC real test set | PSNR: 22.68 -> 22.95 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- BNUDC* — BNUDC: A Two-Branched Deep Neural Network for Restoring Images From Under-Display Cameras (CVPR 2022) · better on 1 of 1 shared result
- TSIT — TSIT: A Simple and Versatile Framework for Image-to-Image Translation (ECCV 2020) · better on 1 of 1 shared result
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
- UDC-SIT: A Real-World Dataset for Under-Display Cameras (NeurIPS 2023 · similarity 0.90)
- Panel-Specific Degradation Representation for Raw Under-Display Camera Image Restoration (ECCV 2024 · similarity 0.87)
- Decoupling Degradations with Recurrent Network for Video Restoration in Under-Display Camera (AAAI 2024 · similarity 0.85)
- Correlation Matching Transformation Transformers for UHD Image Restoration (AAAI 2024 · similarity 0.84)
- Dual Data Alignment Makes AI-Generated Image Detector Easier Generalizable (NeurIPS 2025 · similarity 0.83)
- InstructRestore: Region-Customized Image Restoration with Human Instructions (NeurIPS 2025 · 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.
