J. Daniel Subías - In-the-wild Material Appearance Editing using Perceptual Attributes
J. Daniel Subías · Jornadas de jóvenes investigadores del I3A · 2023
We rely on an encoder-decoder architecture G that encodes the image x, and manipulates the latent space z together with the target attribute att t to generate the edited image y.We use skip connections with STU cells, a variant of the GRU [1,2], which allow encoder-decoder architectures to keep the relevant information of the input image in the edited output when manipulating the latent space z.High-level overview of our framework.Our generator G is composed of an encoder G enc and a decoder G dec .It is capable of editing the input image x according to the target attribute att t to generate the edited image y.