Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Official implementation of the paper "The Stable Signature Rooting Watermarks in Latent Diffusion Models"
| Date | Stars |
|---|---|
| 2026-07-31 | 524 |
| 2026-08-06 | 524 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# ✍️ The Stable Signature: Rooting Watermarks in Latent Diffusion Models Implementation and pretrained models. For details, see [**the paper**](https://arxiv.org/abs/2303.15435) (or go to ICCV 2023 in Paris 🥐). [[`Webpage`](https://pierrefdz.github.io/publications/stablesignature/)] [[`arXiv`](https://arxiv.org/abs/2303.15435)] [[`Blog`](https://ai.meta.com/blog/stable-signature-watermarking-generative-ai/)] [[`Demo`](https://huggingface.co/spaces/imatag/stable-signature-bzh)] ## Setup ### Requirements First, clone the repository locally and move inside the folder: ```cmd git clone https://github.com/facebookresearch/stable_signature cd stable_signature ``` To install the main dependencies, we recommand using conda. [PyTorch](https://pytorch.org/) can be installed with: ```cmd conda install -c pytorch torchvision pytorch==1.12.0 cudatoolkit==11.3 ``` Install the remaining dependencies with pip: ```cmd pip install -r requirements.txt ``` This codebase has been developed with python version 3.8, PyTorch version 1.12.0, CUDA 11.3. ### Models and data #### Data The paper uses the [COCO](https://cocodataset.org/) dataset to fine-tune the LDM decoder (we filtered images containing people). All you need is around 500 images for training (preferably over 256x256). #### Watermark models The watermark extractor model can be downloaded in the following links. The `.pth` file has not been whitened, while the `.torchscript.pt` file has been and can be used without any further processing. We additionally provide another extractor model, which has been trained with blur and rotations and has better robustness to that kind of attacks, at the cost of a slightly lower image quality (you might need to adjust the perceptual loss weight at your convenience). | Model | Checkpoint | Torch-Script | | --- | --- | --- | | Extractor | [dec_48b.pth](https://dl.fbaipublicfiles.com/ssl_watermarking/dec_48b.pth) | [dec_48b_whit.torchscript.pt](https://dl.fbaipublicfiles.com/ssl_watermarking/dec_48b_whit.torchscript.pt) | | Other | [other_dec_48b_whit.pth](https://dl.fbaipublicfiles.com/ssl_watermarking/other_dec_48b.pth) | [other_dec_48b_whit.torchscript.pt](https://dl.fbaipublicfiles.com/ssl_watermarking/other_dec_48b_whit.torchscript.pt) | The following code automatically downloads the models and put them in the `models` folder: ```cmd mkdir models wget https://dl.fbaipublicfiles.com/ssl_watermarking/dec_48b_whit.torchscript.pt -P models/ wget https://dl.fbaipublicfiles.com/ssl_watermarking/other_dec_48b_whit.torchscript.pt -P models/ ``` Code to train the watermark models is available in the folder called `hidden/`. #### Stable Diffusion models Create LDM configs and checkpoints from the [Hugging Face](https://huggingface.co/stabilityai) and [Stable Diffusion](https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion) repositories. The code should also work for Stable Diffusion v1 without any change. For other models (like old LDMs or VQGANs), you may need to adapt the code to load the checkpoints. An example of watermarked weights is available at [WM weights of latent decoder](https://dl.fbaipublicfiles.com/ssl_watermarking/sd2_decoder.pth) (the key is the one present in the `decoding.ipynb` file). #### Perceptual Losses The perceptual losses are based on [this repo](https://github.com/SteffenCzolbe/PerceptualSimilarity/). You should download the weights here: https://github.com/SteffenCzolbe/PerceptualSimilarity/tree/master/src/loss/weights, and put them in a folder called `losses` (this is used in [src/loss/loss_provider.py#L22](https://github.com/facebookresearch/stable_signature/blob/main/src/loss/loss_provider.py#L22)). To do so you can run ``` git clone https://github.com/SteffenCzolbe/PerceptualSimilarity.git cp -r PerceptualSimilarity/src/loss/weights src/loss/losses/ rm -r PerceptualSimilarity ``` ## Usage ### Watermark pre-training Please see [hidden/README.md](https://github.com/facebo
Excerpt of 9,204 characters
Read on GitHub
Pierre Fernandez · Meta · France
18
2
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:914599ec5b3a8b7a, desc:latent diffusion