pierrefdz / stable_signature Goto Github PK

Hi,

I'm getting the following error when I try to run the code:

FileNotFoundError: [Errno 2] No such file or directory: '/checkpoint/pfz/ai_signature/loss_weights/rgb_watson_vgg_trial0.pth'

The command I used was:

 python finetune_ldm_decoder.py --num_keys 1 --ldm_config ./models/768-v-ema.yaml --ldm_ckpt ./models/768-v-ema.ckpt --msg_decoder_path ./models/dec_48b_whit.torchscript.pt --train_dir ./data/coco/images_500_subset/train/train_class --val_dir ./data/coco/images_500_subset/val/val_class

Can you point me to a place to download the required loss weights? Thanks!

Hi,

Thank you for making the code public. If I understand correctly, the provided watermark extractor checkpoint is not the "real" one that is used to report the results in the paper. I was wondering how it compares with the one you used for the paper in terms of performance.

I'm asking because if they are not too different I can start to run some experiments; otherwise I guess I'll have to wait for the release of the "real" extractor? Would appreciate any suggestions or clarifications.

Thanks

Heyho,
I have a problem understanding how to reload the LDM Decoder Weights in Stable Diffusion.

...
config = OmegaConf.load(f"{opt.config}")
model = load_model_from_config(config, f"{opt.ckpt}")

device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
model = model.to(device)

state_dict = torch.load("../stable_signature/output/checkpoint_000.pth")
msg = model.first_stage_model.load_state_dict(state_dict, strict=False)
 ...   

Do I need to add any other lines or reference the msg variable? Because when I generate images with txt2img.py from StabelDiffusion (v1.4) with python scripts/txt2img.py --prompt "Cute tabby cat" --H 256 --W 256 --config "..\stable_signature\sd\stable-diffusion-v-1-4-original\v1-inference.yaml" --ckpt "..\stable_signature\sd\stable-diffusion-v-1-4-original\sd-v1-4-full-ema.ckpt" --n_samples 1 --ddim_steps 32, the code from your Notebook yields an accuracy of 43%.

I used the exact versions and dependencies that the author provided in this repo.

Thanks in advance!