Expand a texture seamlessly
Posted: Sat May 26, 2018 8:49 pm
One month ago this GAN was developed to generate the continuation of a texture seamlessly. The learning process is pretty simple:
The white portion is the original image, the blue portion is a random piece of the original image given as input to the GAN, the red portion will be used as ground truth and will be compared with the guessed output.
This way, instead of letting a texture repeat itself on a surface, we can generate our own continuation for more realistic maps. Since the editor only accepts 256x256 textures, I guess too big textures wouldn't be that cool resized to that size, so we wouldn need to split the big texture and then split the surface in the editor(which Idk how).
The white portion is the original image, the blue portion is a random piece of the original image given as input to the GAN, the red portion will be used as ground truth and will be compared with the guessed output.
This way, instead of letting a texture repeat itself on a surface, we can generate our own continuation for more realistic maps. Since the editor only accepts 256x256 textures, I guess too big textures wouldn't be that cool resized to that size, so we wouldn need to split the big texture and then split the surface in the editor(which Idk how).