Research map / Image generation / Super-resolution
Reference-based Super-Resolution via Image-based Retrieval-Augmented Generation Diffusion
Byeonghun Lee, Hyunmin Cho, Hong Gyu Choi, Soo Min Kang, Iljun Ahn, Kyong Hwan Jin
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 hashing-based reference matching in training (vs random pairing) (ablation) | iRAG · RealSR | PSNR: PSNR from 23.957 to 23.614 without it, SSIM from 0.715 to 0.698 | Improved |
| add the reference restoration module (ablation) | iRAG · RealSR | PSNR: 23.957 -> 23.374 when removed, SSIM 0.715 -> 0.680 | Improved |
| add hashing-based retrieval at inference (vs random patch selection) (ablation) | iRAG · RealSR | PSNR: 23.957 -> 23.709 when removed, SSIM 0.715 -> 0.708 | Improved |
| raise loss-balancing weight alpha from 1.0 to 1.3 (ablation) | iRAG · DIV2K-Val | PSNR vs CLIP-IQA: PSNR 22.323 -> 22.007, CLIP-IQA 0.702 -> 0.716 | Mixed |
| train hash on real plus generated patches instead of real only (ablation) | hash retrieval · CUFED hit rate R1 | hit rate (%): 45 -> 52 | Improved |
| train hash on generated patches only instead of real (ablation) | hash retrieval · CUFED hit rate R1 | hit rate (%): 45 -> 40 | Hurt |
| augment limited real set (1K) with generated patches (ablation) | hash retrieval · CUFED hit rate R1 | hit rate (%): 48 -> 52 | Improved |
| use 16-bit hash codes instead of VGG feature matching | reference retrieval · Retrieval cost | matching time (ms): 4338.2 ms -> 67.8 ms | Improved |
| use hash-selected references instead of generated references | CoSeR · CoSeR reference adaptation | MUSIQ: 56.419 -> 57.686 | Improved |
| use hash references at training and evaluation instead of random (ablation) | TTSR · CUFED | PSNR: 24.74 -> 25.65 | Improved |
Closest papers by meaning
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.91)
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.89)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.89)
- RAP-SR: RestorAtion Prior Enhancement in Diffusion Models for Realistic Image Super-Resolution (AAAI 2025 · similarity 0.88)
- Solving Diffusion ODEs with Optimal Boundary Conditions for Better Image Super-Resolution (ICLR 2024 · similarity 0.88)
- Building Bridges across Spatial and Temporal Resolutions: Reference-Based Super-Resolution via Change Priors and Conditional Diffusion Model (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.
