Research map / 3D vision / Depth and stereo
UnReflectAnything: RGB-Only Highlight Removal by Rendering Synthetic Specular Supervision
Alberto Rota, Mert Kiray, Mert Asim Karaoglu, Patrick Ruhkamp, Elena De Momi, Nassir Navab, Benjamin Busam
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 MoGe-2 normals instead of depth-gradient normals (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | MSE_m: 0.012 -> 0.003 | Improved |
| add MoGe-2 normals (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | SSIM: 0.909 -> 0.957 | Improved |
| add token-space inpainting (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | SSIM: 0.816 -> 0.957 | Improved |
| add token-space inpainting (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | MSE_m: 0.007 -> 0.003 | Improved |
| add local mean prior (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | SSIM: 0.911 -> 0.957 | Improved |
| add local mean prior (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | MSE_m: 0.004 -> 0.003 | No clear effect |
| add decoder pre-training stage (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | SSIM: 0.873 -> 0.957 | Improved |
| add decoder pre-training stage (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | MSE_m: 0.006 -> 0.003 | Improved |
| exclude dataset highlights from supervision (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | MSE_m: 0.022 -> 0.003 | Improved |
| exclude dataset highlights from supervision (ablation) | UnReflectAnything · PSD+SSHR+SHIQ average | SSIM: 0.933 -> 0.957 | Improved |
| UnReflectAnything versus DHAN-SHR | UnReflectAnything · PSD | PSNR: 24.863 -> 17.230 | Hurt |
| UnReflectAnything versus DHAN-SHR | UnReflectAnything · SSHR | PSNR: 37.318 -> 28.415 | Hurt |
Closest papers by meaning
- SHDocs: A dataset, benchmark, and method to efficiently generate high-quality, real-world specular highlight data with near-perfect alignment (NeurIPS 2024 · similarity 0.87)
- Hierarchical Adaptive Filtering Network for Text Image Specular Highlight Removal (CVPR 2025 · similarity 0.85)
- UniLumos: Fast and Unified Image and Video Relighting with Physics-Plausible Feedback (NeurIPS 2025 · similarity 0.85)
- SpecNeRF: Gaussian Directional Encoding for Specular Reflections (CVPR 2024 · similarity 0.85)
- Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation (CVPR 2025 · similarity 0.85)
- Factored-NeuS: Reconstructing Surfaces, Illumination, and Materials of Possibly Glossy Objects (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.
