dailenson / One-DM

Official Code for ECCV 2024 paper — One-Shot Diffusion Mimicker for Handwritten Text Generation
MIT License
298 stars 28 forks source link

Predicted output of the model #3

Open merryTong opened 2 months ago

merryTong commented 2 months ago

Hello, I read your paper and code. The paper describes the model Unet as being used to predict x0, but why is it predicting noise in the code? _17260496019171

QiferG commented 2 months ago

Hello, I read your paper and code. The paper describes the model Unet as being used to predict x0, but why is it predicting noise in the code? ![]()

Thanks to your question, U-Net can predict both noise and x0 directly. There is a conversion relationship between noise ε and x0, making them essentially equivalent. you can get more information here. unet

merryTong commented 2 months ago

I got it. Thanks.