From SRA to Self-Flow: Data Augmentation or Self-Supervision?

Hugging Face Daily Papers Papers

Summary

This paper investigates the mechanisms behind self-alignment methods in diffusion transformers, revealing that performance improvements from methods like Self-Flow primarily come from data augmentation along the noise dimension rather than token interactions between noise levels. The authors introduce Attention Separation to demonstrate this and propose an effective design combining self-representation alignment with dual-timestep augmentation.

Representation alignment has become an effective way to accelerate diffusion transformer training and improve generation quality. Recent self-alignment methods, such as SRA and Self-Flow, further remove the dependency on external pretrained encoders by constructing alignment within the diffusion model itself. However, the mechanism behind the improvement from SRA to Self-Flow, dual-time scheduling, remains under-examined: Self-Flow attributes its gain to interactions between tokens at different noise levels, where cleaner tokens help infer noisier ones. In this work, we revisit this explanation and ask whether the gain instead comes from data augmentation along the noise dimension. To disentangle these factors, we introduce Attention Separation, which preserves the same dual-timestep input as Self-Flow while blocking attention between tokens assigned to different noise levels. Surprisingly, removing such interaction does not degrade performance and can even improve it, suggesting that the improvement from SRA to Self-Flow mainly comes from data augmentation. Furthermore,We show that Attention Separation itself provides an augmentation effect by splitting a single image into multiple effective training parts to expand the training data. Based on these observations, we combine self-representation alignment with dual-timestep and attention-separation augmentation, and demonstrate the effectiveness of this design on ImageNet.
Original Article
View Cached Full Text

Cached at: 07/03/26, 07:53 AM

Paper page - From SRA to Self-Flow: Data Augmentation or Self-Supervision?

Source: https://huggingface.co/papers/2607.02508

Abstract

Research investigates the mechanisms behind self-alignment methods in diffusion transformers, finding that performance improvements stem primarily from data augmentation along the noise dimension rather than token interactions between noise levels.

Representation alignmenthas become an effective way to acceleratediffusion transformertraining and improve generation quality. Recentself-alignmentmethods, such asSRAandSelf-Flow, further remove the dependency on external pretrained encoders by constructing alignment within the diffusion model itself. However, the mechanism behind the improvement fromSRAtoSelf-Flow,dual-time scheduling, remains under-examined:Self-Flowattributes its gain to interactions between tokens at different noise levels, where cleaner tokens help infer noisier ones. In this work, we revisit this explanation and ask whether the gain instead comes fromdata augmentationalong thenoise dimension. To disentangle these factors, we introduceAttention Separation, which preserves the same dual-timestep input asSelf-Flowwhile blocking attention between tokens assigned to different noise levels. Surprisingly, removing such interaction does not degrade performance and can even improve it, suggesting that the improvement fromSRAtoSelf-Flowmainly comes fromdata augmentation. Furthermore,We show thatAttention Separationitself provides an augmentation effect by splitting a single image into multiple effective training parts to expand the training data. Based on these observations, we combine self-representation alignmentwith dual-timestep and attention-separation augmentation, and demonstrate the effectiveness of this design on ImageNet.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.02508

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.02508 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.02508 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.02508 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

LeapAlign: Post-Training Flow Matching Models at Any Generation Step by Building Two-Step Trajectories

Hugging Face Daily Papers

LeapAlign is a post-training method that improves flow matching model alignment with human preferences by reducing computational costs through two-step trajectory shortcuts while enabling stable gradient propagation to early generation steps. The method outperforms state-of-the-art approaches when fine-tuning Flux models across various image quality and text-alignment metrics.