Research map / Image generation / Autoregressive and latent generation
Exploring Position Encoding Mechanism in Diffusion U-Net for Training-free High-resolution Image Generation
Feng Zhou, Pu Cao, Yiyang Ma, Lu Yang, Yonghao Dang, Jianqin Yin
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 direct inference with PBC | SD-XL · 2048x2048 generation | KID: 0.0092 -> 0.0067 | Improved |
| replace direct inference with PBC | SD-XL · 2048x2048 generation | Content Richness: 30.30 -> 29.42 | Improved |
| replace direct inference with PBC | SD-XL · 2048x2048 generation | ImageReward: 0.645 -> 1.036 | Improved |
| replace direct inference with PBC | SD-XL · 2048x2048 generation | Inference time (s): 68s -> 71s (+4%) | Hurt |
| use PBC instead of ScaleCrafter | SD-XL · 2048x2048 generation | IS: 9.53 -> 9.95 | Improved |
| use PBC instead of FouriScale | SD-XL · 2048x2048 generation | Inference time (s): 130s (+91%) -> 71s (+4%) | Improved |
| use PBC instead of ScaleCrafter | SD-XL · 2048x2048 generation | ImageReward: 1.078 -> 1.036 | Hurt |
| add PBC virtual boundaries to zero-padding U-Net (ablation) | SD-2.1 · position loss, 1024^2 | Linear-probe position loss: 0.023 -> 0.017 | Improved |
| add PBC virtual boundaries to zero-padding U-Net (ablation) | SD-2.1 · position loss, central 512^2 in 1024^2 | Linear-probe position loss: 0.039 -> 0.010 | Improved |
| use PBC instead of dilated convolution (ablation) | SD-2.1 · position loss, central 512^2 in 1024^2 | Linear-probe position loss: 0.025 -> 0.010 | Improved |
| add one virtual boundary (N=0 to N=1) (ablation) | PBC · position loss | Linear-probe position loss: 0.0230 -> 0.0173 | Improved |
| increase virtual boundaries from N=1 to N=7 (ablation) | PBC · position loss | Linear-probe position loss: 0.0173 -> 0.0175 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- ScaleCrafter — ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024) · better on 3 of 6 shared results
- FouriScale — FouriScale: A Frequency Perspective on Training-Free High-Resolution Image Synthesis (ECCV 2024) · better on 6 of 6 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
- Native-Resolution Image Synthesis (NeurIPS 2025 · similarity 0.88)
- ScaleCrafter: Tuning-free Higher-Resolution Visual Generation with Diffusion Models (ICLR 2024 · similarity 0.87)
- UltraPixel: Advancing Ultra High-Resolution Image Synthesis to New Peaks (NeurIPS 2024 · similarity 0.87)
- UltraDiff:Transferring High-Fidelity Priors to Compressed Latent Spaces for High-resolution Image Generation (NeurIPS 2026 · similarity 0.87)
- ResMaster: Mastering High-Resolution Image Generation via Structural and Fine-Grained Guidance (AAAI 2025 · similarity 0.87)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 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.
