search

open-mmlab / PIA

[CVPR 2024] PIA, your Personalized Image Animator. Animate your images by text prompt, combing with Dreambooth, achieving stunning videos. PIA，你的个性化图像动画生成器，利用文本提示将图像变为奇妙的动画
https://pi-animator.github.io/
Apache License 2.0
913 stars 75 forks source link

How long to train the model #44

Open ailearnerkyh opened 5 months ago

ailearnerkyh commented 5 months ago

Nice work! How long and how much GPU memory is required to train this model on WebVid-10M?

zengyh1900 commented 3 months ago

Hi @ailearnerkyh, With 8 A100 GPUs, you can have a good checkpoint at about 72k iterations within two days (or maybe 48k within one day). Training longer may lead to better results.