guyyariv/DyPE
Official implementation for "DyPE: Dynamic Position Extrapolation for Ultra High Resolution Diffusion".
Dynamically adjusts positional encodings during the denoising process to match evolving frequency content, enabling pre-trained diffusion transformers to scale to 4K resolution without retraining. Supports multiple position encoding methods (YARN, NTK, base) and integrates with diffusion model architectures including standard transformers and Qwen-Image models. Zero additional sampling overhead compared to baseline inference.
344 stars.
Stars
344
Forks
37
Language
Python
License
MIT
Category
Last pushed
Feb 24, 2026
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/guyyariv/DyPE"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related models
devzhk/InverseBench
InverseBench (ICLR 2025 spotlight)
bahjat-kawar/ddrm
[NeurIPS 2022] Denoising Diffusion Restoration Models -- Official Code Repository
wyhuai/DDNM
[ICLR 2023 Oral] Zero-Shot Image Restoration Using Denoising Diffusion Null-Space Model
yuanzhi-zhu/DiffPIR
"Denoising Diffusion Models for Plug-and-Play Image Restoration", Yuanzhi Zhu, Kai Zhang,...
Algolzw/image-restoration-sde
Image Restoration with Mean-Reverting Stochastic Differential Equations, ICML 2023. Winning...