U-REPA: Aligning Diffusion U-Nets to ViTs

• URL: http://arxiv.org/abs/2503.18414v1
• Date: Mon, 24 Mar 2025 07:46:00 GMT
• Title: U-REPA: Aligning Diffusion U-Nets to ViTs
• Abstract summary: We propose U-REPA, a representation alignment paradigm that bridges U-Net hidden states and ViT features.<n>Experiments indicate that the resulting U-REPA could achieve excellent generation quality and greatly accelerates the convergence speed.
• Score: 29.987838381433445
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Representation Alignment (REPA) that aligns Diffusion Transformer (DiT) hidden-states with ViT visual encoders has proven highly effective in DiT training, demonstrating superior convergence properties, but it has not been validated on the canonical diffusion U-Net architecture that shows faster convergence compared to DiTs. However, adapting REPA to U-Net architectures presents unique challenges: (1) different block functionalities necessitate revised alignment strategies; (2) spatial-dimension inconsistencies emerge from U-Net's spatial downsampling operations; (3) space gaps between U-Net and ViT hinder the effectiveness of tokenwise alignment. To encounter these challenges, we propose U-REPA, a representation alignment paradigm that bridges U-Net hidden states and ViT features as follows: Firstly, we propose via observation that due to skip connection, the middle stage of U-Net is the best alignment option. Secondly, we propose upsampling of U-Net features after passing them through MLPs. Thirdly, we observe difficulty when performing tokenwise similarity alignment, and further introduces a manifold loss that regularizes the relative similarity between samples. Experiments indicate that the resulting U-REPA could achieve excellent generation quality and greatly accelerates the convergence speed. With CFG guidance interval, U-REPA could reach $FID<1.5$ in 200 epochs or 1M iterations on ImageNet 256 $\times$ 256, and needs only half the total epochs to perform better than REPA. Codes are available at https://github.com/YuchuanTian/U-REPA.

Related papers

• UDT: Reconciling U-Nets and Diffusion Transformers with Data-Adaptive Token Reduction [4.511561231517167]
  Diffusion Transformers (DiTs) have emerged as a core architecture in generative modeling.<n>We propose UDT, a U-Net diffusion transformer that combines the representation power of DiTs with the encoding-decoding benefits of U-Nets.
  arXiv  Detail & Related papers  (2026-08-02T15:07:21Z)
• Weaving Light and Time: Unified Harmonic-Geometric Representation Learning for Dense RGB-Event Parsing [55.13802890769086]
  We present Evita, the first unified backbone specifically engineered for dedicated dense RGB-Event parsing.<n>To achieve profound modal synergy, Evita embeds a suite of intrinsic co-learning modules directly into every encoder layer.<n>Evita establishes new state-of-the-art metrics while delivering a superior accuracy-latency trade-off for real-time multimodal perception.
  arXiv  Detail & Related papers  (2026-07-10T06:52:09Z)
• Rethinking the Pointer Loss in Table Structure Recognition: Geometry-Aware Pointer Loss for Spatial Locality [12.986158862036866]
  Table Structure Recognition using a pointer network achieves impressive results by predicting HTML sequences while aligning tags to detected text (or cell) regions.<n>When pointer networks fail, 79.6% of errors occur between spatially adjacent cells.<n>We propose Geometry-Aware Pointer (GAP) Loss, which reweights the cross-entropy objective based on spatial proximity to ground truth.
  arXiv  Detail & Related papers  (2026-06-17T05:59:31Z)
• Improved Baselines with Representation Autoencoders [61.47127824064028]
  Representation Autoencoders (RAE) replace traditional VAE with pretrained vision encoders.<n>We find three insights which simplify and improve RAE.<n> RAEv2 leads to more than 10x faster convergence over the original RAE.
  arXiv  Detail & Related papers  (2026-05-18T12:42:34Z)
• R2G: A Multi-View Circuit Graph Benchmark Suite from RTL to GDSII [14.790942679912595]
  R2G is a multi-view circuit-graph benchmark suite that standardizes five stage-aware views with information parity.<n>R2G provides an end-to-end DEF-to-graph pipeline spanning synthesis, placement, and routing stages.
  arXiv  Detail & Related papers  (2026-04-09T22:59:27Z)
• What matters for Representation Alignment: Global Information or Spatial Structure? [64.67092609921816]
  Representation alignment (REPA) guides generative training by distilling representations from a strong, pretrained vision encoder to intermediate diffusion features.<n>We investigate a fundamental question: what aspect of the target representation matters for generation, its textitglobal revisionsemantic information.<n>We replace the standard projection layer in REPA with a simple convolution layer and introduce a spatial normalization layer for the external representation.
  arXiv  Detail & Related papers  (2025-12-11T16:39:53Z)
• Linear Differential Vision Transformer: Learning Visual Contrasts via Pairwise Differentials [34.77694214755808]
  We introduce Visual-Contrast Attention (VCA), a drop-in replacement for Multi-Head Self-Attention (MHSA)<n>VCA injects an explicit notion of discrimination while reducing the theoretical complexity from O(N N C) to O(N n C) with n N.<n>The module adds fewer than 0.3M parameters to a DeiT-Tiny backbone, requires no extra FLOPs, and is wholly architecture-agnostic.
  arXiv  Detail & Related papers  (2025-11-02T07:04:12Z)
• CoIFNet: A Unified Framework for Multivariate Time Series Forecasting with Missing Values [17.25081407284703]
  Collaborative Imputation-Forecasting Network (CoIFNet) is a novel framework that unifies imputation and forecasting.<n>CoIFNet takes the observed values, mask matrix and timestamp embeddings as input, processing them sequentially.<n>We demonstrate the effectiveness and computational efficiency of our proposed approach across diverse missing-data scenarios.
  arXiv  Detail & Related papers  (2025-06-16T03:15:12Z)
• RT-DETRv3: Real-time End-to-End Object Detection with Hierarchical Dense Positive Supervision [7.721101317599364]
  We propose a hierarchical dense positive supervision method based on RT-DETR, named RT-DETRv3. To address insufficient decoder training, we propose a novel learning strategy involving self-attention perturbation. RT-DETRv3 significantly outperforms existing real-time detectors, including the RT-DETR series and the YOLO series.
  arXiv  Detail & Related papers  (2024-09-13T02:02:07Z)
• U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformers [28.936553798624136]
  We propose a series of U-shaped DiTs (U-DiTs) in the paper and conduct extensive experiments to demonstrate the extraordinary performance of U-DiT models. The proposed U-DiT could outperform DiT-XL/2 with only 1/6 of its cost computation.
  arXiv  Detail & Related papers  (2024-05-04T18:27:29Z)
• TCCT-Net: Two-Stream Network Architecture for Fast and Efficient Engagement Estimation via Behavioral Feature Signals [58.865901821451295]
  We present a novel two-stream feature fusion "Tensor-Convolution and Convolution-Transformer Network" (TCCT-Net) architecture. To better learn the meaningful patterns in the temporal-spatial domain, we design a "CT" stream that integrates a hybrid convolutional-transformer. In parallel, to efficiently extract rich patterns from the temporal-frequency domain, we introduce a "TC" stream that uses Continuous Wavelet Transform (CWT) to represent information in a 2D tensor form.
  arXiv  Detail & Related papers  (2024-04-15T06:01:48Z)
• ELGC-Net: Efficient Local-Global Context Aggregation for Remote Sensing Change Detection [65.59969454655996]
  We propose an efficient change detection framework, ELGC-Net, which leverages rich contextual information to precisely estimate change regions. Our proposed ELGC-Net sets a new state-of-the-art performance in remote sensing change detection benchmarks. We also introduce ELGC-Net-LW, a lighter variant with significantly reduced computational complexity, suitable for resource-constrained settings.
  arXiv  Detail & Related papers  (2024-03-26T17:46:25Z)
• AANet: Aggregation and Alignment Network with Semi-hard Positive Sample Mining for Hierarchical Place Recognition [48.043749855085025]
  Visual place recognition (VPR) is one of the research hotspots in robotics, which uses visual information to locate robots. We present a unified network capable of extracting global features for retrieving candidates via an aggregation module. We also propose a Semi-hard Positive Sample Mining (ShPSM) strategy to select appropriate hard positive images for training more robust VPR networks.
  arXiv  Detail & Related papers  (2023-10-08T14:46:11Z)
• SFNet: Faster and Accurate Semantic Segmentation via Semantic Flow [88.97790684009979]
  A common practice to improve the performance is to attain high-resolution feature maps with strong semantic representation. We propose a Flow Alignment Module (FAM) to learn textitSemantic Flow between feature maps of adjacent levels. We also present a novel Gated Dual Flow Alignment Module to directly align high-resolution feature maps and low-resolution feature maps.
  arXiv  Detail & Related papers  (2022-07-10T08:25:47Z)
• Accelerating DETR Convergence via Semantic-Aligned Matching [50.3633635846255]
  This paper presents SAM-DETR, a Semantic-Aligned-Matching DETR that greatly accelerates DETR's convergence without sacrificing its accuracy. It explicitly searches salient points with the most discriminative features for semantic-aligned matching, which further speeds up the convergence and boosts detection accuracy as well.
  arXiv  Detail & Related papers  (2022-03-14T06:50:51Z)
• Anti-Oversmoothing in Deep Vision Transformers via the Fourier Domain Analysis: From Theory to Practice [111.47461527901318]
  Vision Transformer (ViT) has recently demonstrated promise in computer vision problems. ViT saturates quickly with depth increasing, due to the observed attention collapse or patch uniformity. We propose two techniques to mitigate the undesirable low-pass limitation.
  arXiv  Detail & Related papers  (2022-03-09T23:55:24Z)
• Pruning Self-attentions into Convolutional Layers in Single Path [89.55361659622305]
  Vision Transformers (ViTs) have achieved impressive performance over various computer vision tasks. We propose Single-Path Vision Transformer pruning (SPViT) to efficiently and automatically compress the pre-trained ViTs. Our SPViT can trim 52.0% FLOPs for DeiT-B and get an impressive 0.6% top-1 accuracy gain simultaneously.
  arXiv  Detail & Related papers  (2021-11-23T11:35:54Z)
• RCNet: Reverse Feature Pyramid and Cross-scale Shift Network for Object Detection [10.847953426161924]
  We propose RCNet, which consists of Reverse Feature Pyramid (RevFP) and Cross-scale Shift Network (CSN) RevFP utilizes local bidirectional feature fusion to simplify the bidirectional pyramid inference pipeline. CSN directly propagates representations to both adjacent and non-adjacent levels to enable multi-scale features more correlative.
  arXiv  Detail & Related papers  (2021-10-23T04:00:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.