Weiming Zhang, Nenghai Yu, Han Fang, Kejiang Chen, Kai Zeng, Zijin Yang
We lifted 14 functions out of this paper's own repositories and ran 5 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| bsmhmmlf/Gaussian-Shading | canonical | 5 of 14 |
| Function | Status | Where it lives |
|---|---|---|
| convert_to_custom_text_state_dict | Ran | bsmhmmlf/Gaussian-Shading/open_clip/model.py code served (permissive licence) · get_code("f19962ebb134b3d7") |
| get_cast_dtype | Ran | bsmhmmlf/Gaussian-Shading/open_clip/model.py code served (permissive licence) · get_code("dcd422d66b0581d8") |
| image_distortion | Ran | bsmhmmlf/Gaussian-Shading/image_utils.py code served (permissive licence) · get_code("5fa7b8bcd038a010") |
| latents_to_imgs | Ran | bsmhmmlf/Gaussian-Shading/image_utils.py code served (permissive licence) · get_code("82db872730553ad6") |
| transform_img | Ran | bsmhmmlf/Gaussian-Shading/image_utils.py code served (permissive licence) · get_code("f2eed600b6113eb3") |
| backward_ddim | Not yet run | bsmhmmlf/Gaussian-Shading/inverse_stable_diffusion.py code served (permissive licence) · get_code("b9d521d7021e46bb") |
| build_model_from_openai_state_dict | Not yet run | bsmhmmlf/Gaussian-Shading/open_clip/model.py code served (permissive licence) · get_code("57f1602ed20a9c21") |
| forward_ddim | Not yet run | bsmhmmlf/Gaussian-Shading/inverse_stable_diffusion.py code served (permissive licence) · get_code("3b89ab9f937dccdd") |
| gather_features | Not yet run | bsmhmmlf/Gaussian-Shading/open_clip/loss.py code served (permissive licence) · get_code("ddcbd45e940484ee") |
| hf_read_jsonlines | Not yet run | bsmhmmlf/Gaussian-Shading/io_utils.py code served (permissive licence) · get_code("075fbc6126eeefe6") |
| load_jsonlines | Not yet run | bsmhmmlf/Gaussian-Shading/io_utils.py code served (permissive licence) · get_code("81cef8be4a138b63") |
| prepare_inputs_for_generation | Not yet run | bsmhmmlf/Gaussian-Shading/open_clip/coca_model.py code served (permissive licence) · get_code("fb651d0a97fd4d3f") |
| register_pooler | Not yet run | bsmhmmlf/Gaussian-Shading/open_clip/hf_model.py code served (permissive licence) · get_code("2a377da4a76a2d44") |
| resolve_globs | Not yet run | bsmhmmlf/Gaussian-Shading/io_utils.py code served (permissive licence) · get_code("afbb28c04b32db08") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Ethical concerns surrounding copyright protection and inappropriate content generation pose challenges for the practical implementation of diffusion models. One effective solution involves watermarking the generated images. However, existing methods often compromise the model performance or require additional training, which is undesirable for operators and users. To address this issue, we propose Gaussian Shading, a diffusion model watermarking technique that is both performance-lossless and trainingfree, while serving the dual purpose of copyright protection and tracing of offending content. Our watermark embedding is free of model parameter modifications and thus is plug-and-play. We map the watermark to latent representations following a standard Gaussian distribution, which is indistinguishable from latent representations obtained from the non-watermarked diffusion model. Therefore we can achieve watermark embedding with lossless performance, for which we also provide theoretical proof. Furthermore, since the watermark is intricately linked with image semantics, it exhibits resilience to lossy processing and erasure attempts. The watermark can be extracted by Denoising Diffusion Implicit Models (DDIM) inversion and inverse sampling. We evaluate Gaussian Shading on multiple versions of Stable Diffusion, and the results demonstrate that Gaussian Shading not only is performance-lossless but also outperforms existing methods in terms of robustness.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2404.04956")
get_code_for_paper("2404.04956")
have("2404.04956")
Connect an agent — have() is free.