Research map / Image generation / Super-resolution
IFCSR: Inference-Free Fidelity-Realism Control for One-Step Diffusion-based Real-World Image Super-Resolution
Jonghee Back, Jongju Kim, Jeong-Uk Kim, Eunjin Kim, Minyong Jeon
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 |
|---|---|---|---|
| increase blending parameter gamma from 0 to 1 (image-space blend of fidelity and realism outputs) (ablation) | IFCSR · RealSR (x4) | PSNR: 28.00 -> 22.10 | Hurt |
| increase blending parameter gamma from 0 to 1 (ablation) | IFCSR · RealSR (x4) | CLIPIQA: 0.3337 -> 0.7011 | Improved |
| increase blending parameter gamma from 0 to 1 (ablation) | IFCSR · DRealSR (x4) | PSNR: 30.86 -> 24.63 | Hurt |
| increase blending parameter gamma from 0 to 1 (ablation) | IFCSR · RealSR (x4) | LPIPS: 0.2921 -> 0.2540 -> 0.3501 | Mixed |
| use two-stage training instead of single-stage training with the baseline LPIPS+MSE loss (ablation) | IFCSR · RealSR (x4) | fidelity-realism spectrum alignment | Improved |
| replace baseline LPIPS+MSE loss with depth-weighted specialized losses in two-stage training (ablation) | IFCSR · RealSR (x4) | fidelity-realism spectrum width | Improved |
| use depth-dependent weighting in the feature loss (ablation) | IFCSR · RealSR (x4) | LPIPS monotonicity across gamma | Hurt |
| blend fidelity and realism outputs in image space instead of latent-space control | IFCSR · control cost | extra network inferences per control | Improved |
| blend two outputs with differing structure in image space | IFCSR · failure case | visual artifacts | Hurt |
| generate both fidelity- and realism-specific images with two LoRA networks | IFCSR · Efficiency, RTX 3090 FP16 | inference time (s): 0.20 (PiSA-SR) -> 0.35 | Hurt |
| use IFCSR at gamma = 1.0 instead of PiSA-SR on DRealSR | controllable one-step ISR · DRealSR (x4) | CLIPIQA: 0.6975 -> 0.7446 | Improved |
Closest papers by meaning
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.91)
- Pixel-level and Semantic-level Adjustable Super-resolution: A Dual-LoRA Approach (CVPR 2025 · similarity 0.91)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.90)
- One-Step Effective Diffusion Network for Real-World Image Super-Resolution (NeurIPS 2024 · similarity 0.89)
- DREAM: Diffusion Rectification and Estimation-Adaptive Models (CVPR 2024 · similarity 0.89)
- Unleashing the Power of One-Step Diffusion based Image Super-Resolution via a Large-Scale Diffusion Discriminator (NeurIPS 2025 · similarity 0.89)
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.
