Learnings from Scaling Visual Tokenizers for Reconstruction and Generation

Philippe Hansen-Estruch, David Yan, Ching-Yao Chuang, Orr Zohar, Jialiang Wang, Tingbo Hou, Tao Xu, Sriram Vishwanath, Peter Vajda, Xinlei Chen
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:22023-22043, 2025.

Abstract

Visual tokenization via auto-encoding empowers state-of-the-art image and video generative models by compressing pixels into a latent space. However, questions remain about how auto-encoder design impacts reconstruction and downstream generative performance. This work explores scaling in auto-encoders for reconstruction and generation by replacing the convolutional backbone with an enhanced Vision Transformer for Tokenization (ViTok). We find scaling the auto-encoder bottleneck correlates with reconstruction but exhibits a nuanced relationship with generation. Separately, encoder scaling yields no gains, while decoder scaling improves reconstruction with minimal impact on generation. As a result, we determine that scaling the current paradigm of auto-encoders is not effective for improving generation performance. Coupled with Diffusion Transformers, ViTok achieves competitive image reconstruction and generation performance on 256p and 512p ImageNet-1K. In videos, ViTok achieves SOTA reconstruction and generation performance on 16-frame 128p UCF-101.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-hansen-estruch25a, title = {Learnings from Scaling Visual Tokenizers for Reconstruction and Generation}, author = {Hansen-Estruch, Philippe and Yan, David and Chuang, Ching-Yao and Zohar, Orr and Wang, Jialiang and Hou, Tingbo and Xu, Tao and Vishwanath, Sriram and Vajda, Peter and Chen, Xinlei}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {22023--22043}, 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/hansen-estruch25a/hansen-estruch25a.pdf}, url = {https://proceedings.mlr.press/v267/hansen-estruch25a.html}, abstract = {Visual tokenization via auto-encoding empowers state-of-the-art image and video generative models by compressing pixels into a latent space. However, questions remain about how auto-encoder design impacts reconstruction and downstream generative performance. This work explores scaling in auto-encoders for reconstruction and generation by replacing the convolutional backbone with an enhanced Vision Transformer for Tokenization (ViTok). We find scaling the auto-encoder bottleneck correlates with reconstruction but exhibits a nuanced relationship with generation. Separately, encoder scaling yields no gains, while decoder scaling improves reconstruction with minimal impact on generation. As a result, we determine that scaling the current paradigm of auto-encoders is not effective for improving generation performance. Coupled with Diffusion Transformers, ViTok achieves competitive image reconstruction and generation performance on 256p and 512p ImageNet-1K. In videos, ViTok achieves SOTA reconstruction and generation performance on 16-frame 128p UCF-101.} }
Endnote
%0 Conference Paper %T Learnings from Scaling Visual Tokenizers for Reconstruction and Generation %A Philippe Hansen-Estruch %A David Yan %A Ching-Yao Chuang %A Orr Zohar %A Jialiang Wang %A Tingbo Hou %A Tao Xu %A Sriram Vishwanath %A Peter Vajda %A Xinlei Chen %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-hansen-estruch25a %I PMLR %P 22023--22043 %U https://proceedings.mlr.press/v267/hansen-estruch25a.html %V 267 %X Visual tokenization via auto-encoding empowers state-of-the-art image and video generative models by compressing pixels into a latent space. However, questions remain about how auto-encoder design impacts reconstruction and downstream generative performance. This work explores scaling in auto-encoders for reconstruction and generation by replacing the convolutional backbone with an enhanced Vision Transformer for Tokenization (ViTok). We find scaling the auto-encoder bottleneck correlates with reconstruction but exhibits a nuanced relationship with generation. Separately, encoder scaling yields no gains, while decoder scaling improves reconstruction with minimal impact on generation. As a result, we determine that scaling the current paradigm of auto-encoders is not effective for improving generation performance. Coupled with Diffusion Transformers, ViTok achieves competitive image reconstruction and generation performance on 256p and 512p ImageNet-1K. In videos, ViTok achieves SOTA reconstruction and generation performance on 16-frame 128p UCF-101.
APA
Hansen-Estruch, P., Yan, D., Chuang, C., Zohar, O., Wang, J., Hou, T., Xu, T., Vishwanath, S., Vajda, P. & Chen, X.. (2025). Learnings from Scaling Visual Tokenizers for Reconstruction and Generation. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:22023-22043 Available from https://proceedings.mlr.press/v267/hansen-estruch25a.html.

Related Material