3d-terrain-deep-dream
Pytorch implementation of deep-dreaming on a 3d mesh based on Neural 3D Mesh Renderer
The basic idea with this is that if you have a DEM file with no assosciated textures, that you can use photographs of similar terrain to automatically create fitting textures. This uses a GAN style approach of asking a discriminator to look at real images and rendered images, training the discriminator to get better while the deep dream makes the textures better.
I just started this, so it's not ready to clone yet....
I've tested this using the GeoPose3k Dataset.
Code usage
Usage instructions found here: user manual page.
Example Output
GeoPose3k Dataset
1) Reference Image 2) Generated Texture Render 3) Generated Texture in UV Space
