search

sjtuplayer / few-shot-diffusion

[ICCV 2023] Phasic Content Fusing Diffusion Model with Directional Distribution Consistency for Few-Shot Model Adaption
53 stars 2 forks source link

stop point "Train the whole model on both the source and target dataset" #7

Open fikry102 opened 10 months ago

fikry102 commented 10 months ago

Once it is good enough, you can stop training. In our experience, for the style like cartoon and VanGogh_face, 300-500 iterations are enough while for Sketches, 3000 iterations work better.

"300-500 iterations" corresponds to 500.pth under the folder "output-whole/models"?

image

sjtuplayer commented 10 months ago

yes, you are right.

Rachael-web commented 7 months ago

Once it is good enough, you can stop training. In our experience, for the style like cartoon and VanGogh_face, 300-500 iterations are enough while for Sketches, 3000 iterations work better.

"300-500 iterations" corresponds to 500.pth under the folder "output-whole/models"?

image

But 500.pth under the folder "output-whole/models" has not even completed the training of one epoch. Is this okay?

15634960802 commented 4 months ago

Once it is good enough, you can stop training. In our experience, for the style like cartoon and VanGogh_face, 300-500 iterations are enough while for Sketches, 3000 iterations work better. “300-500 次迭代”对应于文件夹“output-whole/models”下的 500.pth？ image

但是文件夹“output-whole/models”下的 500.pth 甚至还没有完成一个纪元的训练。这样可以吗？

我也很奇怪啊