Research map / Image generation / Super-resolution
LIIFusion: Coarse-to-fine Framework for Generative MEF via Implicit Neural Representation
Sangmin Han, Jinho Kim, Jinwoo Kim, Dongyoung Kim, Seon Joo 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 |
|---|---|---|---|
| replace patch-wise diffusion at HR with low-res diffusion plus LIIF fine stage | UltraFusion · UltraFusion Benchmark | total inference time: 203 min -> 59 min | Improved |
| coarse-to-fine design reduces compute | UltraFusion · 1988x1326 image | TFLOPs: 944.4 -> 282.1 | Improved |
| LIIFusion vs UltraFusion on perceptual quality | UltraFusion · UltraFusion Benchmark | MUSIQ/HyperIQA: MUSIQ 68.40 -> 70.19; HyperIQA 0.6214 -> 0.6467 | Improved |
| LIIFusion vs UltraFusion on DeQA-Score | UltraFusion · UltraFusion Benchmark | DeQA-Score: 4.0123 -> 3.9807 | Hurt |
| LIIFusion vs UltraFusion on structural MEF-SSIM | UltraFusion · MEFB | MEF-SSIM: 0.9266 -> 0.9201 | Hurt |
| add adaptive exposure correction (alpha 0.2) (ablation) | coarse stage · MEFB | MEF-SSIM: 0.9266 -> 0.9313 | Improved |
| add adaptive exposure correction (alpha 0.2) (ablation) | coarse stage · MEFB | MUSIQ: 68.14 -> 67.91 | Hurt |
| raise AEC alpha to 1.0 (ablation) | coarse stage · MEFB | MEF-SSIM: 0.9313 -> 0.8633 | Hurt |
| train LIIF for multi-exposure fusion instead of using SR-pretrained LIIF (ablation) | fine stage · MEFB | MEF-SSIM, HyperIQA | Improved |
| feed both OE and UE inputs to fine stage (ablation) | fine stage · UltraFusion Benchmark | NRIQA metrics | Improved |
| train on UltraFusion pseudo labels instead of SICE labels (ablation) | LIIFusion · UltraFusion Benchmark | NRIQA metrics | No clear effect |
| add ICCV23 pseudo-labeled data to training (ablation) | LIIFusion · UltraFusion Benchmark | MUSIQ/PAQ2PIQ/HyperIQA: MUSIQ 70.19 -> 70.22; PAQ2PIQ 75.59 -> 75.46; HyperIQA 0.6467 -> 0.6506 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UltraFusion — UltraFusion: Ultra High Dynamic Imaging using Exposure Fusion (CVPR 2025) · better on 9 of 12 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.86)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.85)
- Fourier Priors-Guided Diffusion for Zero-Shot Joint Low-Light Enhancement and Deblurring (CVPR 2024 · similarity 0.85)
- ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024 · similarity 0.85)
- RaPD: Resolution-Agnostic Pixel Diffusion via Semantics-Enriched Implicit Representations (NeurIPS 2026 · similarity 0.85)
- AGLLDiff: Guiding Diffusion Models Towards Unsupervised Training-free Real-world Low-light Image Enhancement (AAAI 2025 · similarity 0.85)
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.
