Research map / Image generation / Super-resolution
Enhanced Latent-Space Adversarial Training for Super-Resolution
Liangbin Xie, Zheyuan Li, Fanghua Yu, Xinqi Lin, Jun-hao Zhuang, Jinfan Hu, Jinjin Gu, Jiantao Zhou, Chao Dong
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 |
|---|---|---|---|
| remove degradation removal encoder and noise augmentation (ablation) | HYPIR cascade · RealPhoto60 x4 | over-saturation (qualitative) | Improved |
| move adversarial loss from image space to FLUX latent space (ablation) | HYPIR++ · RealPhoto60 | NIQE: 5.957 -> 4.155 | Improved |
| add latent patch U-Net discriminator (ablation) | HYPIR++ latent-space training · RealPhoto60 | NIQE / DeQA: NIQE 4.155 -> 3.977; DeQA 4.209 -> 4.247 | Improved |
| add latent patch U-Net discriminator (ablation) | HYPIR++ latent-space training · RealPhoto60 | MUSIQ: 74.712 -> 74.382 | Hurt |
| shorten text tokens from 512 to 128 (ablation) | HYPIR++ · RealPhoto60 1024x1024 | inference time (s): 3.44 -> 3.06 (1.12x) | Improved |
| shorten text tokens from 512 to 128 (ablation) | HYPIR++ · RealPhoto60 512x512 | inference time (s): 1.20 -> 0.90 (1.33 ×) | Improved |
| replace full attention with sparse neighbor attention (SNA) (ablation) | HYPIR++ x4 cascade · RealPhoto60 x4 | inference time (s): 16.12 -> 9.42 (1.71x) | Improved |
| replace full attention with SNA (ablation) | HYPIR++ x4 cascade · RealPhoto60 x4 | MUSIQ: 73.913 -> 73.440 | Hurt |
| replace HYPIR with HYPIR++ in two-stage cascade | HYPIR · RealSR x8 and DRealSR x8 | NIQE, MANIQA, CLIP-IQA, DeQA, MUSIQ: RealSR DeQA 3.837 -> 4.028 | Improved |
| replace HYPIR with HYPIR++ | HYPIR · RealPhoto60 x4 | MUSIQ vs OSEDiff: HYPIR++ 73.44 vs OSEDiff 74.73 | Mixed |
| HYPIR++ versus HYPIR in user study | HYPIR · user study, 20 image groups | preference share: 13.2% -> 86.8% | Improved |
Closest papers by meaning
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · similarity 0.87)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.86)
- Adversarial Diffusion Compression for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.86)
- Iterative Token Evaluation and Refinement for Real-World Super-resolution (AAAI 2024 · similarity 0.86)
- Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · similarity 0.85)
- Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution (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.
