Research map / Image generation / Super-resolution
Toward Bridging the Simulated-to-Real Gap: Benchmarking Super-Resolution on Real Data
Thomas Köhler, Michel Bätz, Farzad Naderi, André Kaup, Andreas Maier, Christian Rieß
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 |
|---|---|---|---|
| Evaluate on real hardware-binned LR data instead of simulated bicubic LR data | 19 SR algorithms (SISR and MFSR) · SupER global motion, 2x2 and 4x4 binning | Spearman rank correlation between simulated and real rankings: rho < 0.6 | Hurt |
| Use deep networks (VDSR, DRCN) instead of classical SR methods on real data | SISR · SupER real data | PSNR/SSIM/IFC rank | No clear effect |
| Use multi-frame instead of single-image SR under pure global motion | SR algorithms · SupER global motion | normalized PSNR and IFC | Improved |
| Use multi-frame instead of single-image SR under mixed motion | SR algorithms · SupER mixed motion | normalized PSNR and IFC | Hurt |
| Use multi-frame instead of single-image SR under local motion | SR algorithms · SupER local motion | normalized PSNR and IFC | Hurt |
| Use external training data instead of self-exemplars | SISR · SupER global motion | normalized PSNR and IFC | Improved |
| Add H.265/HEVC compression to LR input (QP up to 40) | all SR algorithms · SupER global motion, 3x | normalized PSNR and IFC | Hurt |
| Add photometric outlier frames to the input set | MFSR · SupER global motion, 3x, K=11 frames | normalized PSNR and IFC | Hurt |
| Use reconstruction-based robust models (IRWSR, BEPSR) instead of interpolation under photometric variations | MFSR · SupER photometric variation | normalized PSNR and IFC | Improved |
| Increase magnification factor in observer study | Observer agreement · SupER observer study | Kendall coefficient of agreement u | Improved |
| Omit sanity checks (tolerate all 8 errors, n_f = 8) (ablation) | Observer study · SupER observer study | inter-observer variance: n_f = 8 vs n_f = 1 | Hurt |
| Use LPIPS or IFC instead of PSNR/SSIM/MS-SSIM as quality measure | Full-reference measures · SupER vs B-T observer scores | weighted Kendall tau distance | Improved |
Closest papers by meaning
- Rethinking Image Super Resolution from Training Data Perspectives (ECCV 2024 · similarity 0.90)
- SEAL: A Framework for Systematic Evaluation of Real-World Super-Resolution (ICLR 2024 · similarity 0.90)
- Investigating Tradeoffs in Real-World Video Super-Resolution (CVPR 2022 · similarity 0.89)
- Sequence Matters: Harnessing Video Models in 3D Super-Resolution (AAAI 2025 · similarity 0.89)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.88)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.88)
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.
