2 ms·
These AI enhancers just make the image look like whatever the training set looked like. My favorite example of how flawed it is https://video-images.vice.com/te
by Gigachad 4y ago
These AI enhancers just make the image look like whatever the training set looked like. My favorite example of how flawed it is https://video-images.vice.com/test-uploads/articles/5ef216e9aea1330097d9ec64/lede/1592924226088-obama.jpeg?crop=1xw:0.7835555555555556xh;center,center https://video-images.vice.com/test-uploads/articles/5ef216e9...
There is really no solution to this because even we couldn't tell this is wrong without prior knowledge of the left side.
- mcphage 4y agoThere was one issue I remember reading about where when someone upscale a picture, the upscale stick Ryan Reynold’s face in a building randomly.
- mawise 4y agoWhats the source of that image? It looks like a suspicious result, for example the white shirt is lost in the "upscaled" image. Was is trained on a specific set of images to prove this point?
- potatoman22 4y agoIt's from an algorithm called PULSE. It seems to search the latent space of StyleGAN to find a face that compresses to something similar to the query image. https://arxiv.org/pdf/2003.03808.pdf https://arxiv.org/pdf/2003.03808.pdf https://www.theverge.com/21298762/face-depixelizer-ai-machine-learning-tool-pulse-stylegan-obama-bias https://www.theverge.com/21298762/face-depixelizer-ai-machin...
- mawise 4y agoThanks for the link!