Research map / Image generation / Super-resolution
Any-Resolution Training for High-Resolution Image Synthesis
Lucy Chai, Michaël Gharbi, Eli Shechtman, Phillip Isola, Richard 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 scale conditioning (ablation) | inverse-teacher Anyres-GAN · FFHQ | FID@1024: 4.88 -> 4.47 | Improved |
| add scale conditioning (ablation) | inverse-teacher Anyres-GAN · FFHQ | pFID: 4.67 -> 4.28 | Improved |
| use inverse instead of forward teacher loss (ablation) | Anyres-GAN · FFHQ6K | FID@1024: 5.35 -> 4.88 | Improved |
| add teacher regularization (inverse) (ablation) | Anyres-GAN · FFHQ6K | FID@1024: 7.13 -> 4.88 | Improved |
| add teacher regularization (inverse) (ablation) | Anyres-GAN · FFHQ6K | pFID: 3.17 -> 4.67 | Hurt |
| raise teacher weight from 2 to 5 (ablation) | Anyres-GAN · FFHQ6K | FID@1024 and pFID: FID@1024 5.46 -> 4.47; pFID 3.15 -> 4.28 | Mixed |
| multisize training instead of resizing all images down to 512 (ablation) | Anyres-GAN · FFHQ | FID@1024: 19.18 -> 4.47 | Improved |
| multisize training instead of upsampling all images to 1024 (ablation) | Anyres-GAN · FFHQ | FID@1024: 4.86 -> 4.47 | Improved |
| multisize training instead of 1024 subset only (ablation) | Anyres-GAN · FFHQ | FID@512: 12.41 -> 4.41 | Improved |
| increase HR images from 1k to 70k (ablation) | Anyres-GAN · FFHQ | FID: 4.38 -> 4.52 | No clear effect |
| sample training scales from (512, 1024) instead of (256, 512) (ablation) | Anyres-GAN · FFHQ6K | FID@1024: 19.01 -> 4.16 | Improved |
| remove base-resolution sampling for the discriminator (ablation) | Anyres-GAN · FFHQ6K | FID@256: 3.28 -> 9.96 | Hurt |
Closest papers by meaning
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.90)
- UltraPixel: Advancing Ultra High-Resolution Image Synthesis to New Peaks (NeurIPS 2024 · similarity 0.90)
- Native-Resolution Image Synthesis (NeurIPS 2025 · similarity 0.90)
- ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024 · similarity 0.89)
- ResMaster: Mastering High-Resolution Image Generation via Structural and Fine-Grained Guidance (AAAI 2025 · similarity 0.89)
- GSDD: Generative Space Dataset Distillation for Image Super-resolution (AAAI 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.
