Learning-to-Cache: A Novel Method for Accelerating Diffusion Transformers Through Layer Caching
The Learning-to-Cache (L2C) method accelerates the inference of diffusion transformers by dynamically learning which layers can be cached and reused across timesteps without retraining, leading to significant speedups with minimal impact on image quality.
SmoothCache: A Training-Free Method for Accelerating Inference in Diffusion Transformer Models Across Modalities
SmoothCache is a novel, training-free technique that significantly speeds up inference in Diffusion Transformer models across various modalities (image, video, audio) by intelligently caching and reusing similar layer outputs from adjacent diffusion timesteps, achieving performance comparable to or exceeding existing methods without compromising generation quality.
Truncated Consistency Models: A Two-Stage Training Approach for Improved Diffusion Model Generation
Truncated Consistency Models (TCM) improve the efficiency and sample quality of diffusion models by focusing training on the latter stages of the generative process, thereby allocating more network capacity to generation rather than denoising.
加速擴散模型推理：重新思考編碼器角色以實現更快的擴散
本文核心論點為，透過分析擴散模型中 UNet 編碼器和解碼器的特徵演變，發現編碼器特徵在多個時間步長中變化極小，而解碼器特徵則表現出顯著變化。基於此發現，作者提出編碼器傳播方法，透過在相鄰時間步長中重複使用編碼器特徵，實現高效的擴散採樣，並在保持圖像品質的同時顯著減少 UNet 和基於 Transformer 的擴散模型在多種生成任務上的推理時間。
Faster Diffusion Model Inference: Omitting Encoder Computations for Parallel Decoding
Diffusion model inference can be significantly accelerated by omitting encoder computations at certain time steps and reusing previously computed encoder features, enabling parallel decoding without sacrificing image quality.
Improving Diffusion Model Sampling Efficiency with Optimal Covariance Matching: A Method for Learning Diagonal Covariances from Score Functions
This paper introduces Optimal Covariance Matching (OCM), a novel method for enhancing the sampling efficiency of diffusion models by learning the diagonal covariances of denoising distributions directly from score functions, leading to improved generation quality, recall rate, and likelihood estimation.
고해상도 Diffusion 모델의 효율성을 위한 딥 압축 오토인코더
고해상도 이미지 합성에 널리 사용되는 Latent Diffusion 모델의 속도를 높이기 위해, 본 논문에서는 높은 공간 압축률을 가진 새로운 오토인코더인 DC-AE를 제안하며, 잔여 오토인코딩 및 분리된 고해상도 적응 기술을 통해 기존 오토인코더보다 뛰어난 성능과 효율성을 달성했습니다.
Deep Compression Autoencoder (DC-AE): A Novel Approach for Accelerating High-Resolution Diffusion Models
This paper introduces DC-AE, a new type of autoencoder that significantly speeds up high-resolution image synthesis in diffusion models by achieving higher spatial compression ratios while maintaining reconstruction accuracy.
基於有限差分逼近策略的免訓練自適應擴散模型
本文提出了一種名為 AdaptiveDiffusion 的新型 diffusion 模型加速方法，該方法可以根據輸入提示自適應地減少去噪過程中的噪聲預測步驟，從而在保持生成質量的同時顯著提高效率。
경계 차분 근사 전략을 사용한 훈련 없는 적응형 Diffusion 모델
Diffusion 모델의 노이즈 예측 단계를 입력 프롬프트에 따라 적응적으로 줄여 계산 비용을 줄이면서도 생성 결과의 품질을 유지하는 AdaptiveDiffusion이라는 새로운 접근 방식을 제안합니다.
© 2024 by Linnk AI