Get the latest tech news
Imagine Flash: Accelerating Emu Diffusion Models with Backward Distillation
Diffusion models are a powerful generative framework, but come with expensive inference. Existing acceleration methods often compromise image quality or...
Our approach comprises three key components: (i) Backward Distillation, which mitigates training-inference discrepancies by calibrating the student on its own backward trajectory; (ii) Shifted Reconstruction Loss that dynamically adapts knowledge transfer based on the current time step; and (iii) Noise Correction, an inference time technique that enhances sample quality by addressing singularities in noise prediction. Armen Avetisyan, Chris Xie, Henry Howard-Jenkins, Tsun-Yi Yang, Samir Aroudj, Suvam Patra, Fuyang Zhang, Duncan Frost, Luke Holland, Campbell Orme, Jakob Julian Engel, Edward Miller, Richard Newcombe, Vasileios Balntas Felix Xu, Di Lin, Jianjun Zhao, Jianlang Chen, Lei Ma, Qing Guo, Wei Feng, Xuhong Ren
Or read this on Hacker News
