Research map / Image generation / Autoregressive and latent generation
DyPE: Dynamic Position Extrapolation for Ultra High Resolution Diffusion
Noam Issachar, Guy Yariv, Sagie Benaim, Yossi Adi, Dani Lischinski, Raanan Fattal
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 |
|---|---|---|---|
| make YaRN time-dependent (DY-YaRN) | YaRN on FLUX · DrawBench 4096x4096 | ImageReward: -0.34 -> 0.15 | Improved |
| make YaRN time-dependent (DY-YaRN) | YaRN on FLUX · Aesthetic-4K 4096x4096 | FID: 192.19 -> 186.00 | Improved |
| make NTK-aware time-dependent (DY-NTK) | NTK on FLUX · DrawBench 4096x4096 | ImageReward: -1.88 -> -1.22 | Improved |
| make NTK-aware time-dependent (DY-NTK) | NTK on FLUX · Aesthetic-4K 4096x4096 | FID: 203.85 -> 183.72 | Improved |
| make YaRN time-dependent at low resolution | YaRN on FLUX · Aesthetic-4K 2048x3072 | FID: 179.54 -> 179.51 | No clear effect |
| use DY-NTK instead of time-aware scaled RoPE (TASR) | TASR on FLUX · Aesthetic-4K 4096x4096 | FID: 221.39 -> 183.72 | Improved |
| apply DY-YaRN at 6144x6144 | DY-YaRN on FLUX · resolution scaling, ImageReward | ImageReward | Hurt |
| make YaRN time-dependent (DY-YaRN) | YaRN on FiTv2-XL/2 · ImageNet 384x384 | FID: 22.63 -> 21.75 | Improved |
| make PI time-dependent (DY-PI) | PI on FiTv2-XL/2 · ImageNet 384x384 | FID: 118.60 -> 39.56 | Improved |
| make NTK time-dependent (DY-NTK) | NTK on FiTv2-XL/2 · ImageNet 384x384 | FID: 36.75 -> 36.04 | Improved |
| add HiFlow on top of DY-YaRN | DY-YaRN · 4096x4096 non-PE comparison | FID: 186.00 -> 169.46 | Improved |
| add HiFlow on top of DY-YaRN | DY-YaRN · 4096x4096 non-PE comparison | patch-FID: 78.33 -> 79.64 | 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 1 on most shared results.
- HiFlow — HiFlow: Training-free High-Resolution Image Generation with Flow-Aligned Guidance (NeurIPS 2025) · better on 4 of 4 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
- Dynamic Diffusion Transformer (ICLR 2025 · similarity 0.91)
- D^2iT: Dynamic Diffusion Transformer for Accurate Image Generation (CVPR 2025 · similarity 0.91)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.89)
- Effective Diffusion Transformer Architecture for Image Super-Resolution (AAAI 2025 · similarity 0.89)
- SEGA: Spectral-Energy Guided Attention for Resolution Extrapolation in Diffusion Transformers (NeurIPS 2026 · similarity 0.89)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.89)
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.
