optml-group / unlearn-saliency Goto Github PK

Hello~

I notice that the training script for erasing nudity is not given in SD/README.md. Could you share it?

Besides, to repoduce the results for erasing nudity, I encountered some problems:

1. generate mask for nudity involves a dataset named "data/nsfw". How to get this dataset?
2. Saliency unlearning needs remain_dl for training. How to get the remain data for erasing nudity?

Thank you very much~😁

Hi, what an amazing work!
I am trying to reproduce the algorithm recently but I encounter some problems. I use the weights you provided in SD/README.md and it is a file ends with ".ckpt". I try to load it but the keys are incorrect with SDv1.4 no matter I use torch.load(ckpt) or torch.load(ckpt)['state_dict']. Then I use the convertModels.py to convert it into "xx.pt". I use extract_ema=True but it produces exactly same weights as the original SD unet. So I use extract_ema=False and it successfully produces edited weights. However, when I generate I2P images with this weights, the results is far worse than the results in your paper.
So could you help me address this issue? Thanks for your reply!

Hi,

I really liked your work!!!
Could you clarify the GPU requirements needed to train the models?

Thanks,
Kartik

Hi there,

Thank you for sharing your great work! I have a quick question regarding the training of CIFAR10 conditional diffusion. Based on the code:

CUDA_VISIBLE_DEVICES="0,1" python train.py --config cifar10_train.yml --mode train

does not yield a complete conditional model on all 10 classes. Is my understanding correct?

If so, could the authors verify that their results of CIFAR-10 class-wise unlearning in the paper are unaffected and based on the correct implementation.