REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Leng, Xingjian, Singh, Jaskirat, Hou, Yunzhong, Xing, Zhenchang, Xie, Saining, Zheng, Liang
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912663380951040
author Leng, Xingjian
Singh, Jaskirat
Hou, Yunzhong
Xing, Zhenchang
Xie, Saining
Zheng, Liang
author_facet Leng, Xingjian
Singh, Jaskirat
Hou, Yunzhong
Xing, Zhenchang
Xie, Saining
Zheng, Liang
contents In this paper we tackle a fundamental question: "Can we train latent diffusion models together with the variational auto-encoder (VAE) tokenizer in an end-to-end manner?" Traditional deep-learning wisdom dictates that end-to-end training is often preferable when possible. However, for latent diffusion transformers, it is observed that end-to-end training both VAE and diffusion-model using standard diffusion-loss is ineffective, even causing a degradation in final performance. We show that while diffusion loss is ineffective, end-to-end training can be unlocked through the representation-alignment (REPA) loss -- allowing both VAE and diffusion model to be jointly tuned during the training process. Despite its simplicity, the proposed training recipe (REPA-E) shows remarkable performance; speeding up diffusion model training by over 17x and 45x over REPA and vanilla training recipes, respectively. Interestingly, we observe that end-to-end tuning with REPA-E also improves the VAE itself; leading to improved latent space structure and downstream generation performance. In terms of final performance, our approach sets a new state-of-the-art; achieving FID of 1.12 and 1.69 with and without classifier-free guidance on ImageNet 256 x 256. Code is available at https://end2end-diffusion.github.io.
format Preprint
id arxiv_https___arxiv_org_abs_2504_10483
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers
Leng, Xingjian
Singh, Jaskirat
Hou, Yunzhong
Xing, Zhenchang
Xie, Saining
Zheng, Liang
Computer Vision and Pattern Recognition
Machine Learning
In this paper we tackle a fundamental question: "Can we train latent diffusion models together with the variational auto-encoder (VAE) tokenizer in an end-to-end manner?" Traditional deep-learning wisdom dictates that end-to-end training is often preferable when possible. However, for latent diffusion transformers, it is observed that end-to-end training both VAE and diffusion-model using standard diffusion-loss is ineffective, even causing a degradation in final performance. We show that while diffusion loss is ineffective, end-to-end training can be unlocked through the representation-alignment (REPA) loss -- allowing both VAE and diffusion model to be jointly tuned during the training process. Despite its simplicity, the proposed training recipe (REPA-E) shows remarkable performance; speeding up diffusion model training by over 17x and 45x over REPA and vanilla training recipes, respectively. Interestingly, we observe that end-to-end tuning with REPA-E also improves the VAE itself; leading to improved latent space structure and downstream generation performance. In terms of final performance, our approach sets a new state-of-the-art; achieving FID of 1.12 and 1.69 with and without classifier-free guidance on ImageNet 256 x 256. Code is available at https://end2end-diffusion.github.io.
title REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2504.10483