Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers

Katherine Crowson, Stefan Andreas Baumann, Alex Birch, Tanishq Mathew Abraham, Daniel Z Kaplan, Enrico Shippole
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:9550-9575, 2024.

Abstract

We present the Hourglass Diffusion Transformer (HDiT), an image-generative model that exhibits linear scaling with pixel count, supporting training at high resolution (e.g. $1024 \times 1024$) directly in pixel-space. Building on the Transformer architecture, which is known to scale to billions of parameters, it bridges the gap between the efficiency of convolutional U-Nets and the scalability of Transformers. HDiT trains successfully without typical high-resolution training techniques such as multiscale architectures, latent autoencoders or self-conditioning. We demonstrate that HDiT performs competitively with existing models on ImageNet $256^2$, and sets a new state-of-the-art for diffusion models on FFHQ-$1024^2$. Code is available at https://github.com/crowsonkb/k-diffusion.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-crowson24a, title = {Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers}, author = {Crowson, Katherine and Baumann, Stefan Andreas and Birch, Alex and Abraham, Tanishq Mathew and Kaplan, Daniel Z and Shippole, Enrico}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {9550--9575}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/crowson24a/crowson24a.pdf}, url = {https://proceedings.mlr.press/v235/crowson24a.html}, abstract = {We present the Hourglass Diffusion Transformer (HDiT), an image-generative model that exhibits linear scaling with pixel count, supporting training at high resolution (e.g. $1024 \times 1024$) directly in pixel-space. Building on the Transformer architecture, which is known to scale to billions of parameters, it bridges the gap between the efficiency of convolutional U-Nets and the scalability of Transformers. HDiT trains successfully without typical high-resolution training techniques such as multiscale architectures, latent autoencoders or self-conditioning. We demonstrate that HDiT performs competitively with existing models on ImageNet $256^2$, and sets a new state-of-the-art for diffusion models on FFHQ-$1024^2$. Code is available at https://github.com/crowsonkb/k-diffusion.} }
Endnote
%0 Conference Paper %T Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers %A Katherine Crowson %A Stefan Andreas Baumann %A Alex Birch %A Tanishq Mathew Abraham %A Daniel Z Kaplan %A Enrico Shippole %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-crowson24a %I PMLR %P 9550--9575 %U https://proceedings.mlr.press/v235/crowson24a.html %V 235 %X We present the Hourglass Diffusion Transformer (HDiT), an image-generative model that exhibits linear scaling with pixel count, supporting training at high resolution (e.g. $1024 \times 1024$) directly in pixel-space. Building on the Transformer architecture, which is known to scale to billions of parameters, it bridges the gap between the efficiency of convolutional U-Nets and the scalability of Transformers. HDiT trains successfully without typical high-resolution training techniques such as multiscale architectures, latent autoencoders or self-conditioning. We demonstrate that HDiT performs competitively with existing models on ImageNet $256^2$, and sets a new state-of-the-art for diffusion models on FFHQ-$1024^2$. Code is available at https://github.com/crowsonkb/k-diffusion.
APA
Crowson, K., Baumann, S.A., Birch, A., Abraham, T.M., Kaplan, D.Z. & Shippole, E.. (2024). Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass Diffusion Transformers. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:9550-9575 Available from https://proceedings.mlr.press/v235/crowson24a.html.

Related Material