No other representation component is needed: Diffusion transformers can provide representation guidance by themselves
Rating
4 - Good
Authors
Dengyang Jiang, Mengmeng Wang
Date
2025
Review Status
Todo
Review Date
2026/03/10 02:25
Key Findings
Venue
Field
Diffusion
URL
https://arxiv.org/pdf/2505.02831
Paper Library
R
Review Type
Summary
•
Method
◦
external representation model 없이, EMA 모델의 higher layer feature를 target feature alignment으로 사용하여 훈련
•
Result
◦
초반 수렴은 REPA가 빠르고, 최종 수렴에서는 유사한 성능 달성
