I tried using ``AdverseCleaner'' that can remove hostile noise that interferes with AI learning from images
Mist .' A tool called AdverseCleaner has been introduced to remove noise caused by this adversarial sample from images.
In order to prevent the image generation AI from learning illustrations and photographs, there is a way to add noise called adversarial samples to the image using something like '
GitHub - lllyasviel/AdverseCleaner: Remove adversarial noise from images
AdverseCleaner is published on Hugging Face and you can try it out for yourself.
AdverseCleaner - a Hugging Face Space by p1atdev
This time, we will process the following image with adversarial sample noise added using AdverseCleaner.
Load the image and click 'Start'.
The image with noise removed is displayed on the right.
Below is a comparison of the image with noise (left) and the image with noise removed (right). You can compare them by moving the slider in the center. In the image with noise, a unique pattern stands out in the background and painted areas, but the pattern clearly disappears in the image after noise removal.
If you do not have images with noise added, you can also check the effect of AdverseCleaner with the three sample images provided on Hugging Face. Click on the image in 'Examples' and it will be filled in automatically.
a bilateral filter and a guided filter , and Advanced Config adjusts the arguments of these two filters.
Click 'Advanced Config' under Start to fine-tune the noise removal settings. AdverseCleaner consists of
Move the slide bar and click 'Start' to output the image after noise removal.
The image with noise (left) and the image with noise removed (right) are below. You can clearly see that the noise seen in the skin, clothes, and background has been removed and is now clean.
AdverseCleaner is published on HuggingFace and is also distributed as an extension of the AUTOMATIC1111 version Stable Diffusion web UI .
GitHub - gogodr/AdverseCleanerExtension: Remove adversarial noise from images