|国家预印本平台
首页|Upsample What Matters: Region-Adaptive Latent Sampling for Accelerated Diffusion Transformers

Upsample What Matters: Region-Adaptive Latent Sampling for Accelerated Diffusion Transformers

Upsample What Matters: Region-Adaptive Latent Sampling for Accelerated Diffusion Transformers

来源:Arxiv_logoArxiv
英文摘要

Diffusion transformers have emerged as an alternative to U-net-based diffusion models for high-fidelity image and video generation, offering superior scalability. However, their heavy computation remains a major obstacle to real-world deployment. Existing acceleration methods primarily exploit the temporal dimension such as reusing cached features across diffusion timesteps. Here, we propose Region-Adaptive Latent Upsampling (RALU), a training-free framework that accelerates inference along spatial dimension. RALU performs mixed-resolution sampling across three stages: 1) low-resolution denoising latent diffusion to efficiently capture global semantic structure, 2) region-adaptive upsampling on specific regions prone to artifacts at full-resolution, and 3) all latent upsampling at full-resolution for detail refinement. To stabilize generations across resolution transitions, we leverage noise-timestep rescheduling to adapt the noise level across varying resolutions. Our method significantly reduces computation while preserving image quality by achieving up to 7.0$\times$ speed-up on FLUX and 3.0$\times$ on Stable Diffusion 3 with minimal degradation. Furthermore, RALU is complementary to existing temporal accelerations such as caching methods, thus can be seamlessly integrated to further reduce inference latency without compromising generation quality.

Wongi Jeong、Kyungryeol Lee、Hoigi Seo、Se Young Chun

计算技术、计算机技术

Wongi Jeong,Kyungryeol Lee,Hoigi Seo,Se Young Chun.Upsample What Matters: Region-Adaptive Latent Sampling for Accelerated Diffusion Transformers[EB/OL].(2025-07-11)[2025-08-02].https://arxiv.org/abs/2507.08422.点此复制

评论