Research map / Image generation / Autoregressive and latent generation
PRISM: Latent Composition Consistency for Single-Image Reflection Removal
Junseong Shin, Tae Hyun Kim
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 cycle loss L_cycle (ablation) | PRISM · Real(20) | PSNR: 26.67 -> 26.90 | Improved |
| add L_LCS (ablation) | PRISM · Real(20) | PSNR: 26.67 -> 26.74 | Improved |
| add both L_cycle and L_LCS (ablation) | PRISM · Avg PSNR of Real/Nature/SIR2 (unweighted) | PSNR: 27.20 -> 27.48 | Improved |
| supervise both transmission and reflection in cycle branch (ablation) | PRISM · Avg PSNR of Real/Nature/SIR2 (unweighted) | PSNR: 27.34 -> 27.48 | Improved |
| use patch size 4 instead of global average pooling in LCS (ablation) | PRISM · Avg PSNR of Real/Nature/SIR2 (unweighted) | PSNR: 27.29 -> 27.48 | Improved |
| use patch size 4 instead of 8 in LCS (ablation) | PRISM · Avg PSNR of Real/Nature/SIR2 (unweighted) | PSNR: 27.33 -> 27.48 | Improved |
| use patch size 4 instead of 2 in LCS (ablation) | PRISM · Avg PSNR of Real/Nature/SIR2 (unweighted) | PSNR: 27.44 -> 27.48 | Improved |
| add LCC+LCS to SD-2.1 backbone (ablation) | SD-2.1 U-Net · Sample-weighted avg PSNR over 494 images (Real/Nature/SIR2) | PSNR: 24.34 -> 24.73 | Improved |
| add LCC+LCS to SD-2.1 backbone on Nature (ablation) | SD-2.1 U-Net · Nature(20) | PSNR: 26.56 -> 26.40 | Hurt |
| use FLUX backbone instead of SD-2.1 with LCC+LCS (ablation) | PRISM · Sample-weighted avg PSNR over 494 images (Real/Nature/SIR2) | PSNR: 24.73 -> 27.90 | Improved |
| add stop-gradient to cycle loss (ablation) | PRISM · Sample-weighted avg PSNR over 494 images (Real/Nature/SIR2) | PSNR: 27.78 -> 27.90 | Improved |
| fix t=0 instead of uniform t sampling (ablation) | PRISM · Sample-weighted avg PSNR over 494 images (Real/Nature/SIR2) | PSNR: 25.27 -> 27.90 | Improved |
Closest papers by meaning
- Reflection Separation from a Single Image via Joint Latent Diffusion (CVPR 2026 · similarity 0.90)
- Rectifying Latent Space for Generative Single-Image Reflection Removal (CVPR 2026 · similarity 0.90)
- ReflexSplit: Single Image Reflection Separation via Layer Fusion-Separation (CVPR 2026 · similarity 0.88)
- Single Image Reflection Separation via Component Synergy (ICCV 2023 · similarity 0.88)
- Flash-Split: 2D Reflection Removal with Flash Cues and Latent Diffusion Separation (CVPR 2025 · similarity 0.86)
- FIRM: Flexible Interactive Reflection ReMoval (AAAI 2025 · similarity 0.86)
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.
