Paper
adamian98/pulse
krantirk/Self-Supervised-Photo
b5071/pulse-rev
yrsolo/pulse
deeplearningnapratica/pulse
"Explore self-supervised photo upsampling by exploring the latent space of generative models"
-
PUSLE:Photo Upsampling via Latent Space Exploration简称。
-
The purpose of single image super-resolution is to construct a high-resolution (HR) image from the corresponding low-resolution (LR) input.
-
This work is mainly to convert fuzzy low-resolution images into clear, realistic high-resolution images.
-
For a given input LR image, the manifold parameterized by the latent space of the generative model is traversed to find the area with the correct scale. Our method always produces a solution that is both on the natural graphics manifold and can be correctly downsampled to the original resolution image, so it can provide a series of interesting high resolution possibilities.
-
Time-consuming
Each image is generated in about 5 seconds on a single NVIDIA V100 GPU.
- PULSE made imaginary faces of people who do not exist, which should not be confused with real people.