Zhendong-Wang / Prompt-Diffusion

Official PyTorch implementation of the paper "In-Context Learning Unlocked for Diffusion Models"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

training epoch

Jasonhyw opened this issue · comments

Hi,
How many epoch did your checkpoint train ?Thanks.

We finetune Prompt Diffusion for 5000 steps, nearly 20 epochs, with batch size 256*4=1024.

We finetune Prompt Diffusion for 5000 steps, nearly 20 epochs, with batch size 256*4=1024.

I trained it for 9999 steps, nearly 100 epochs, with batch size 172*4=688. But when I valued it, the resulting image is a pink noise image.I don't know the reason. May you have any advice? Thanks.

The reason could be the datasets. I am not sure what dataset you use. I got confirmed from others the code could reproduce the original results with Instruct Pix2Pix modified data.