GFP-GAN, or Generative Facial Prior, is an artificial intelligence-powered tool that can restore and enhance old images. The source image can be low quality and still generate impressive results. How so? It doesn’t simply use a pre-trained StyleGAN and then re-train it to orient the encoded information for their task as PULSE does.
That’s right, it utilizes a pre-trained StyleGAN-2 model to orient their own generative model at multiple scales during the encoding of the image, right down to the latent code and up to reconstruction. Simply put as possible, instead of orientating the training on just the generated (fake) image and the expected (real) image using a discriminator model from the GAN network, GFP-GAN also has two metrics for preserving identity as well as facial components. Speaking of identity, AI can also show us what cartoon characters look like as real-life humans.
- Small but Powerful: KAMRUI Essenx E1 N97 Mini PC Desktop about 3.96× 3.96× 1.57 Inch. Powered with Intel Processor N97 (max up to 3.6GHz, 4C/4T, 6MB...
- Intelligent Design: What sets this small computer desktop apart from other mini PCs is that it comes with 1 HDMI port and 1 DisplayPort, making it...
- Ready When You Are: Just plug-in device is running and ready for you to use simply by opening it up. Equipped with UHD Graphics 1.2 GHz support 4K UHD...
The facial component loss is basically the same thing as the discriminator adversarial loss we find in classic GANs but focuses on important local features of the resulting image like eyes and mouth. The identity preserving loss uses a pre-trained face recognition model to capture the most important facial features and compare them to the real image to see if we still have the same person in the generated image,” said Louis Bouchard.
