Research map / Image generation / Super-resolution
Perceive, Understand and Restore: Real-World Image Super-Resolution with Autoregressive Multimodal Generative Models
Hongyang Wei, Shuaizheng Liu, Chun Yuan, Lei Zhang
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 degradation-level perception guidance (ablation) | PURE · DrealSR (2x) | PSNR: 23.43 -> 24.48 | Improved |
| add semantic description guidance only (ablation) | PURE · DrealSR (2x) | FID: 191.21 -> 141.05 | Improved |
| use both degradation perception and semantic understanding (full guidance) vs perception only (ablation) | PURE · DrealSR (2x) | FID: 83.20 -> 79.41 | Improved |
| replace fixed Top-1 sampling with entropy-based dynamic Top-k (ablation) | PURE · DrealSR (2x) | FID: 115.39 -> 79.41 | Improved |
| replace fixed Top-1 with dynamic Top-k (ablation) | PURE · DrealSR (2x) | PSNR: 27.59 -> 24.73 | Hurt |
| replace fixed Top-2000 with dynamic Top-k (ablation) | PURE · DrealSR (2x) | FID: 80.54 -> 79.41 | Improved |
| use PURE (autoregressive MLLM with VQGAN) instead of SD-based diffusion methods | Real-ISR · RealSR (4x) | PSNR: 25.15 -> 22.83 | Hurt |
| use PURE instead of SD-based methods on no-reference metrics | Real-ISR · RealLQ250 | TOPIQ: 0.7629 -> 0.7751 | Improved |
| use VQGAN tokenizer instead of VAE | Real-ISR reconstruction upper bound · DrealSR (2x) | PSNR: 36.20 -> 33.59 | Hurt |
| use autoregressive MLLM instead of one-step diffusion | OSEDiff vs PURE · 512x512 input | inference time (s): 0.35 -> 256.87 | Hurt |
Closest papers by meaning
- NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution (NeurIPS 2026 · similarity 0.91)
- SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024 · similarity 0.90)
- Visual Autoregressive Modeling for Image Super-Resolution (ICML 2025 · similarity 0.90)
- The Power of Context: How Multimodality Improves Image Super-Resolution (CVPR 2025 · similarity 0.89)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.88)
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · 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.
