LinCa: Accelerating Diffusion Models via Learnable Decomposed Feature Caching
LinCa:基于可学习分解特征缓存的扩散模型加速方法
机构 * Shanghai Jiao Tong University(上海交通大学) ; Shandong University(山东大学) ; Terminal Intelligent Computing Division, Alibaba Cloud(阿里云终端智能计算事业部) ; South China University of Technology(华南理工大学) ; Jilin University(吉林大学)
专题命中 视频扩散模型 :video generation(abstract);分类 cs.CV
AI总结 本文提出LinCa框架,通过可学习可逆网络分解缓存特征并差异化预测,仅需少量额外参数即可在5-7倍加速下使扩散模型保持近无损质量,性能优于现有方法。
Comments Accepted to ECCV 2026. 28 pages including appendix. Code: this https URL (https://github.com/QHR69/LinCa)