UltraPixel: Advancing Ultra-High-Resolution Image Synthesis to New Peaks
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913416260616192 |
|---|---|
| author | Ren, Jingjing Li, Wenbo Chen, Haoyu Pei, Renjing Shao, Bin Guo, Yong Peng, Long Song, Fenglong Zhu, Lei |
| author_facet | Ren, Jingjing Li, Wenbo Chen, Haoyu Pei, Renjing Shao, Bin Guo, Yong Peng, Long Song, Fenglong Zhu, Lei |
| contents | Ultra-high-resolution image generation poses great challenges, such as increased semantic planning complexity and detail synthesis difficulties, alongside substantial training resource demands. We present UltraPixel, a novel architecture utilizing cascade diffusion models to generate high-quality images at multiple resolutions (\textit{e.g.}, 1K to 6K) within a single model, while maintaining computational efficiency. UltraPixel leverages semantics-rich representations of lower-resolution images in the later denoising stage to guide the whole generation of highly detailed high-resolution images, significantly reducing complexity. Furthermore, we introduce implicit neural representations for continuous upsampling and scale-aware normalization layers adaptable to various resolutions. Notably, both low- and high-resolution processes are performed in the most compact space, sharing the majority of parameters with less than 3$\%$ additional parameters for high-resolution outputs, largely enhancing training and inference efficiency. Our model achieves fast training with reduced data requirements, producing photo-realistic high-resolution images and demonstrating state-of-the-art performance in extensive experiments. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_02158 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | UltraPixel: Advancing Ultra-High-Resolution Image Synthesis to New Peaks Ren, Jingjing Li, Wenbo Chen, Haoyu Pei, Renjing Shao, Bin Guo, Yong Peng, Long Song, Fenglong Zhu, Lei Computer Vision and Pattern Recognition Ultra-high-resolution image generation poses great challenges, such as increased semantic planning complexity and detail synthesis difficulties, alongside substantial training resource demands. We present UltraPixel, a novel architecture utilizing cascade diffusion models to generate high-quality images at multiple resolutions (\textit{e.g.}, 1K to 6K) within a single model, while maintaining computational efficiency. UltraPixel leverages semantics-rich representations of lower-resolution images in the later denoising stage to guide the whole generation of highly detailed high-resolution images, significantly reducing complexity. Furthermore, we introduce implicit neural representations for continuous upsampling and scale-aware normalization layers adaptable to various resolutions. Notably, both low- and high-resolution processes are performed in the most compact space, sharing the majority of parameters with less than 3$\%$ additional parameters for high-resolution outputs, largely enhancing training and inference efficiency. Our model achieves fast training with reduced data requirements, producing photo-realistic high-resolution images and demonstrating state-of-the-art performance in extensive experiments. |
| title | UltraPixel: Advancing Ultra-High-Resolution Image Synthesis to New Peaks |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2407.02158 |