Research map / Image generation / Text-to-image diffusion
One-Step Specular Highlight Removal with Adapted Diffusion Models
Mahir Atmis, Levent Karacan, Mehmet Sarıgül
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 |
|---|---|---|---|
| use ProbLoRA-LS instead of standard LoRA (ablation) | SD-Turbo one-step model · SHIQ | PSNR: 33.935 -> 36.190 | Improved |
| use ProbLoRA-LS instead of standard LoRA (ablation) | SD-Turbo one-step model · SSHR | PSNR: 34.117 -> 35.763 | Improved |
| apply ProbLoRA soft selection to linear layers only rather than all layers (ablation) | ProbLoRA · SHIQ | PSNR: 34.752 -> 36.190 | Improved |
| use soft weighting instead of hard top-k row selection (ablation) | ProbLoRA on linear layers · SHIQ | PSNR: 33.648 -> 36.190 | Improved |
| use soft weighting instead of hard top-k row selection (ablation) | ProbLoRA on linear layers · SSHR | PSNR: 35.986 -> 35.763 | Mixed |
| set temperature tau to 0.7 instead of 0.5 (ablation) | ProbLoRA · SHIQ | PSNR: 33.664 -> 36.190 | Improved |
| set temperature tau to 0.7 instead of 1.0 (ablation) | ProbLoRA · SHIQ | PSNR: 33.275 -> 36.190 | Improved |
| use L1 loss instead of L2 loss (ablation) | one-step diffusion model · SHIQ | PSNR: 34.381 -> 35.329 | Improved |
| use multi-scale L1 instead of single-scale L1 (ablation) | one-step diffusion model · SHIQ | PSNR: 35.329 -> 35.383 | Improved |
| add perceptual loss to multi-scale L1 (ablation) | one-step diffusion model · SHIQ | PSNR: 35.383 -> 36.190 | Improved |
| use ProbLoRA instead of LoRA for Fill50K image generation | ControlNet-style generation · Fill50K | FID: 10.384 -> 6.788 | Improved |
| replace multi-stage specular highlight removal with one-step adapted SD-Turbo | TSHRNet · SSHR | PSNR: 28.90 -> 35.76 | Improved |
Closest papers by meaning
- Hierarchical Adaptive Filtering Network for Text Image Specular Highlight Removal (CVPR 2025 · similarity 0.89)
- PHR-DIFF: Portrait Highlights Removal via Patch-aware Diffusion Model (AAAI 2025 · similarity 0.88)
- 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)
- Detail-Preserving Latent Diffusion for Stable Shadow Removal (CVPR 2025 · similarity 0.86)
- Diff-2-in-1: Bridging Generation and Dense Perception with Diffusion Models (ICLR 2025 · similarity 0.85)
- Diffusion Renderer: Neural Inverse and Forward Rendering with Video Diffusion Models (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.
