Research map / Image generation / Image restoration
RevealLayer: Disentangling Hidden and Visible Layers via Occlusion-Aware Image Decomposition
Binhao Wang, Shihao Zhao, Bo Cheng, Qiuyu Ji, Yuhang Ma, Liebucha Wu, Shanyuan Liu, Dawei Leng, Yuhui Yin
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 Region-Aware Attention (RAA) (ablation) | RevealLayer baseline · RevealLayer-100K background | PSNR: 22.08 -> 23.80 | Improved |
| add RAA (ablation) | RevealLayer baseline · RevealLayer-100K foreground | FID: 20.86 -> 19.66 | Improved |
| add Occlusion-Guided Adapter (OGA) to RAA (ablation) | RevealLayer · RevealLayer-100K background | PSNR: 23.80 -> 24.97 | Improved |
| add orthogonality loss (ablation) | RAA + OGA · RevealLayer-100K background | FID: 57.51 -> 53.63 | Improved |
| add orthogonality loss (ablation) | RAA + OGA · RevealLayer-100K foreground | SoftIoU: 0.9271 -> 0.9418 | Improved |
| add alpha loss (ablation) | RAA + OGA · RevealLayer-100K foreground | SoftIoU: 0.9271 -> 0.9420 | Improved |
| add alpha loss to orthogonality-loss variant (ablation) | RAA + OGA + OrthLoss · RevealLayer-100K background | PSNR: 24.96 -> 25.53 | Improved |
| share the same noise across foreground layers at inference (ablation) | RevealLayer · RevealLayerBench foreground | PSNR: 32.13 -> 32.31 | Improved |
| 5%-10% offset in input bounding box (ablation) | RevealLayer · RevealLayerBench foreground | SoftIoU: 0.9432 -> 0.8569 | Hurt |
| enlarge input box by 10%-20% (ablation) | RevealLayer · RevealLayerBench background | PSNR: 25.53 -> 25.30 | Hurt |
| fine-tune VAE decoder on RevealLayer-100K (XVAE) vs original ART VAE (ablation) | RevealLayer · natural-scene background reconstruction | PSNR: 46.68 -> 48.74 | Improved |
| fine-tune RevealLayer 4k steps on PrismLayers vs CLD | RevealLayer · PrismLayersPro validation | PSNR / FID: PSNR 27.646 -> 28.360; FID 19.413 -> 22.716 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OmniPaint — OmniPaint: Mastering Object-Oriented Editing via Disentangled Insertion-Removal Inpainting (ICCV 2025) · better on 4 of 5 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
- Cycle-Consistent Tuning for Layered Image Decomposition (CVPR 2026 · similarity 0.87)
- Referring Layer Decomposition (ICLR 2026 · similarity 0.87)
- Object-Driven Multi-Layer Scene Decomposition From a Single Image (ICCV 2019 · similarity 0.87)
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.87)
- O^2-Recon: Completing 3D Reconstruction of Occluded Objects in the Scene with a Pre-trained 2D Diffusion Model (AAAI 2024 · similarity 0.85)
- Generative Image Layer Decomposition with Visual Effects (CVPR 2025 · similarity 0.84)
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.
