TL; DR:本文提出一种两阶段的扩散模型蒸馏方法,第一阶段将 CFG 蒸馏到模型内部,避免在 CFG 文生图时执行两次模型推理;第二阶段使用类似 Progressive Distillation 的方法进行蒸馏,降低推理所需步数,从两方面提高文本引导的扩散模型的生图速度。 方法 本文提出的蒸馏方法分为两个阶段,第一阶段是对 CFG 进行蒸馏,将 ...
On Distillation of Guided Diffusion Models Chenlin Meng, Ruiqi Gao, Diederik P. Kingma, Stefano Ermon, Jonathan Ho, Tim Salimans arXiv 2022. [Paper] 6 Oct 2022 Improving Sample Quality of Diffusion Model Using Self-Attention Guidance Susung Hong, Gyuseong Lee, Wooseok Jang, Seungryong...
时序建模探索:在之前的方法中,都是利用像素级的扩散,而 MagicVideo 则是最早采用潜在扩散模型(Latent Diffusion Model,LDM)来进行潜在空间的 T2V 生成的工作之一。通过在低维潜在空间中利用扩散模型,它显著降低了计算复杂性,从而加快了处理速度。引入的逐帧轻量级 adaptor 对齐了图像和视频的分布,使所提出的定向注意...
Dataset distillationis the task of synthesizing a small dataset such that models trained on it achieve high performance on the original large dataset. A dataset distillation algorithm takes asinputa large real dataset to be distilled (training set), andoutputsa small synthetic distilled dataset, whic...
title: Diffusion Models Beat GANs on Image Synthesis accepted: NeurIPS 2021 paper: https://arxiv.org/abs/2105.05233 code: https://github.com/openai/guided-diffusion ref: https://sunlin-ai.github.io/2022/05/30/guided-diffusion.html ref: https://blog.csdn.net/NGUever15/article/details/128269...
This paper examines a mathematical diffusion model for the purification of substances by distillation. We present the dependences of the separation factor—a quantitative characteristic of the degree of purification of substances—and the effective separation coefficient on the Peclet number, vaporization ...
During local training, we design a transfer learning (TL) method based on MoE that constructs relatively customized models according to the comprehensive opinions of each expert for different participants. The MoE utilizes a gated network to assign the contributions of various experts through a set ...
TaxDiff: taxonomic-guided diffusion model for protein sequence generation Lin Z Y, Li H, Lv L Z H, et al Sci China Inf Sci, 2025, 68(4): 149101Keywords: controlled protein sequence generation; protein sequence design; diffusion model; generative models; protein sequence generationCite as: Li...
I stand by the post, including the two biggest takeaways that I highlighted (emergent chain-of-thought via pure reinforcement learning, and the power of distillation), and I mentioned the low cost (which I expanded on in Sharp Tech) and chip ban implications, but those observations were too...
2025 ESWA Difussion Ueco: Unified editing chain for efficient appearance transfer with multimodality-guided diffusion - 2025 WACV Difussion Realistic and Efficient Face Swapping: A Unified Approach with Diffusion Models Code 2024 arXiv Diffusion HiFiVFS: High Fidelity Video Face Swapping - 2024 arXiv...