search

zideliu / StyleDrop-PyTorch

Unoffical implement for [StyleDrop](https://arxiv.org/abs/2306.00983)
MIT License
552 stars 26 forks source link

About generate watermark in image #3

Open xyxxmb opened 1 year ago

xyxxmb commented 1 year ago

Hi, Author. There is always a certain probability that a watermark will be generated in the middle of the image when using the trained model to generate it. What is the reason for this? I would be very grateful if you could answer.

zideliu commented 1 year ago

The base model is trained on the CC3M dataset containing watermarked data

xyxxmb commented 1 year ago

Thanks, in this case, I want to use assets/ckpts/imagenet256-450000.ckpt for training, I just modify config.resume_root in configs/custom.py, but it doesn't seem to work. Can you provide a entire configuration file that supports continue training from assets/ckpts/imagenet256-450000.ckpt?

zideliu commented 1 year ago

For this work, we just train an adapter, if you want to resume from the .ckpt, you can get what you want in README_MUSE.md