Scaling Laws in Patchification: An Image Is Worth 50,176 Tokens And More

Feng Wang, Yaodong Yu, Wei Shao, Yuyin Zhou, Alan Yuille, Cihang Xie
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:65278-65290, 2025.

Abstract

Since the introduction of Vision Transformer (ViT), patchification has long been regarded as a common image pre-processing approach for plain visual architectures. By compressing the spatial size of images, this approach can effectively shorten the token sequence and reduce the computational cost of ViT-like plain architectures. In this work, we aim to thoroughly examine the information loss caused by this patchification-based compressive encoding paradigm and how it affects visual understanding. We conduct extensive patch size scaling experiments and excitedly observe an intriguing scaling law in patchification: the models can consistently benefit from decreased patch sizes and attain improved predictive performance, until it reaches the minimum patch size of 1*1, i.e., pixel tokenization. This conclusion is broadly applicable across different vision tasks, various input scales, and diverse architectures such as ViT and the recent Mamba models. Moreover, as a by-product, we discover that with smaller patches, task-specific decoder heads become less critical for dense prediction. In the experiments, we successfully scale up the visual sequence to an exceptional length of 50,176 tokens, achieving a competitive test accuracy of 84.6% with a base-sized model on the ImageNet-1k benchmark. We hope this study can provide insights and theoretical foundations for future works of building non-compressive vision models.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-wang25ed, title = {Scaling Laws in Patchification: An Image Is Worth 50,176 Tokens And More}, author = {Wang, Feng and Yu, Yaodong and Shao, Wei and Zhou, Yuyin and Yuille, Alan and Xie, Cihang}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {65278--65290}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/wang25ed/wang25ed.pdf}, url = {https://proceedings.mlr.press/v267/wang25ed.html}, abstract = {Since the introduction of Vision Transformer (ViT), patchification has long been regarded as a common image pre-processing approach for plain visual architectures. By compressing the spatial size of images, this approach can effectively shorten the token sequence and reduce the computational cost of ViT-like plain architectures. In this work, we aim to thoroughly examine the information loss caused by this patchification-based compressive encoding paradigm and how it affects visual understanding. We conduct extensive patch size scaling experiments and excitedly observe an intriguing scaling law in patchification: the models can consistently benefit from decreased patch sizes and attain improved predictive performance, until it reaches the minimum patch size of 1*1, i.e., pixel tokenization. This conclusion is broadly applicable across different vision tasks, various input scales, and diverse architectures such as ViT and the recent Mamba models. Moreover, as a by-product, we discover that with smaller patches, task-specific decoder heads become less critical for dense prediction. In the experiments, we successfully scale up the visual sequence to an exceptional length of 50,176 tokens, achieving a competitive test accuracy of 84.6% with a base-sized model on the ImageNet-1k benchmark. We hope this study can provide insights and theoretical foundations for future works of building non-compressive vision models.} }
Endnote
%0 Conference Paper %T Scaling Laws in Patchification: An Image Is Worth 50,176 Tokens And More %A Feng Wang %A Yaodong Yu %A Wei Shao %A Yuyin Zhou %A Alan Yuille %A Cihang Xie %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-wang25ed %I PMLR %P 65278--65290 %U https://proceedings.mlr.press/v267/wang25ed.html %V 267 %X Since the introduction of Vision Transformer (ViT), patchification has long been regarded as a common image pre-processing approach for plain visual architectures. By compressing the spatial size of images, this approach can effectively shorten the token sequence and reduce the computational cost of ViT-like plain architectures. In this work, we aim to thoroughly examine the information loss caused by this patchification-based compressive encoding paradigm and how it affects visual understanding. We conduct extensive patch size scaling experiments and excitedly observe an intriguing scaling law in patchification: the models can consistently benefit from decreased patch sizes and attain improved predictive performance, until it reaches the minimum patch size of 1*1, i.e., pixel tokenization. This conclusion is broadly applicable across different vision tasks, various input scales, and diverse architectures such as ViT and the recent Mamba models. Moreover, as a by-product, we discover that with smaller patches, task-specific decoder heads become less critical for dense prediction. In the experiments, we successfully scale up the visual sequence to an exceptional length of 50,176 tokens, achieving a competitive test accuracy of 84.6% with a base-sized model on the ImageNet-1k benchmark. We hope this study can provide insights and theoretical foundations for future works of building non-compressive vision models.
APA
Wang, F., Yu, Y., Shao, W., Zhou, Y., Yuille, A. & Xie, C.. (2025). Scaling Laws in Patchification: An Image Is Worth 50,176 Tokens And More. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:65278-65290 Available from https://proceedings.mlr.press/v267/wang25ed.html.

Related Material