Research map / Image generation / Super-resolution
Fidelity- and Perception-Aware Local Implicit Attention for Arbitrary-Scale Image Super-Resolution
Yu-Syuan Xu, Hao-Lun Sun, Hao-Wei Chen, Hsien-Kai Kuo, Chun-Yi Lee
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 FPAM cross-feature attention (ablation) | FPLIA · DIV2K val x4 | LPIPS: 0.231 -> 0.188 | Improved |
| add FPSM per-pixel selection (ablation) | FPLIA · DIV2K val x4 | LPIPS: 0.231 -> 0.211 | Improved |
| combine FPAM and FPSM (ablation) | FPLIA · DIV2K val x4 | PSNR/LPIPS: 28.40 -> 28.91 PSNR; 0.231 -> 0.183 LPIPS | Improved |
| add cross-attention candidates f_fp and f_pf to f_f + f_p (ablation) | FPAM · DIV2K val x4 | LPIPS: 0.206 -> 0.183 | Improved |
| add perceptual stream f_p to fidelity feature only (ablation) | FPAM · DIV2K val x4 | PSNR: 29.51 -> 28.71 | Hurt |
| per-pixel selection instead of concatenation (ablation) | FPSM · DIV2K val x8 | LPIPS: 0.371 -> 0.338 | Improved |
| per-pixel selection instead of per-channel selection (ablation) | FPSM · DIV2K val x8 | LPIPS: 0.366 -> 0.338 | Improved |
| Spatial Feature Transform instead of per-pixel selection (ablation) | FPSM · DIV2K val x8 | LPIPS: 0.338 -> 0.375 | Hurt |
| FPLIA vs best regression method | FPLIA · Set14 x4 | LPIPS: LPIPS 0.267 -> 0.188; PSNR 29.10 -> 28.48 | Mixed |
| FPLIA vs Kim et al. diffusion method | FPLIA · Urban100 x4 | PSNR/LPIPS: 26.42 -> 26.87 PSNR; 0.170 -> 0.160 LPIPS | Improved |
| FPLIA vs Kim et al. diffusion method | FPLIA · Set5 x4..x16 | LPIPS: x4 0.128 -> 0.116; x8 0.284 -> 0.238 | Mixed |
| add FPLIA modules over Kim et al. pipeline | FPLIA · 128->512 latency | seconds: 94.0 -> 98.4 s; 25.1 -> 25.5 GB | Mixed |
Closest papers by meaning
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · similarity 0.88)
- Pixel-level and Semantic-level Adjustable Super-resolution: A Dual-LoRA Approach (CVPR 2025 · similarity 0.88)
- Auto-Encoded Supervision for Perceptual Image Super-Resolution (CVPR 2025 · similarity 0.87)
- DP²O-SR: Direct Perceptual Preference Optimization for Real-World Image Super-Resolution (NeurIPS 2025 · similarity 0.87)
- Self-Adaptive Reality-Guided Diffusion for Artifact-Free Super-Resolution (CVPR 2024 · similarity 0.87)
- SRSR: Enhancing Semantic Accuracy in Real-World Image Super-Resolution with Spatially Re-Focused Text-Conditioning (NeurIPS 2025 · similarity 0.87)
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.
