Research map / Image generation / Image restoration
Neural Solver of Dichromatic Reflection Model for Specular Highlight Removal
Gang Fu
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 HDNet to ACENet (ablation) | ACENet · SHIQ | PSNR: 33.041 -> 33.843 | Improved |
| add HDNet to ACENet (ablation) | ACENet · SHIQ | SSIM: 0.958 -> 0.967 | Improved |
| add RNet to ACENet+HDNet (ablation) | ACENet+HDNet · SHIQ | PSNR: 33.843 -> 34.501 | Improved |
| add RNet to ACENet+HDNet (ablation) | ACENet+HDNet · SHIQ | SSIM: 0.967 -> 0.979 | Improved |
| supervise D and R rather than intermediate variables (ablation) | Neural DRM solver · SHIQ | PSNR: 33.705 -> 34.501 | Improved |
| supervise D and R rather than intermediate variables (ablation) | Neural DRM solver · SHIQ | SSIM: 0.963 -> 0.979 | Improved |
| add CCFGB guidance to ACENet (ablation) | Neural DRM solver · SHIQ | PSNR: 33.974 -> 34.501 | Improved |
| extend Q-induced to Q-K-induced attention in CCFGB (ablation) | CCFGB · SHIQ | PSNR: 34.380 -> 34.429 | Improved |
| extend Q-K-induced to Q-K-V-induced attention in CCFGB (ablation) | CCFGB · SHIQ | PSNR: 34.429 -> 34.501 | Improved |
| raise HDNet/ACENet input resolution from 50p to 100p (ablation) | Neural DRM solver · SHIQ | PSNR: 34.501 -> 34.513 | No clear effect |
| raise HDNet/ACENet input resolution from 50p to 200p (ablation) | Neural DRM solver · SHIQ | FLOPs (G): 43.57 -> 74.08 | Hurt |
| lower HDNet/ACENet input resolution from 50p to 25p (ablation) | Neural DRM solver · SHIQ | PSNR: 34.501 -> 34.053 | Hurt |
Closest papers by meaning
- Hierarchical Adaptive Filtering Network for Text Image Specular Highlight Removal (CVPR 2025 · similarity 0.87)
- SHDocs: A dataset, benchmark, and method to efficiently generate high-quality, real-world specular highlight data with near-perfect alignment (NeurIPS 2024 · similarity 0.85)
- Diffusion Reflectance Map: Single-Image Stochastic Inverse Rendering of Illumination and Reflectance (CVPR 2024 · similarity 0.85)
- Reti-Diff: Illumination Degradation Image Restoration with Retinex-based Latent Diffusion Model (ICLR 2025 · similarity 0.84)
- PHR-DIFF: Portrait Highlights Removal via Patch-aware Diffusion Model (AAAI 2025 · similarity 0.84)
- Learning to Remove Wrinkled Transparent Film with Polarized Prior (CVPR 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.
