Physically based rendering texture maps using CNNs.
(This is my first ML project so it may be messy and bad.)
For 3D creation, sometimes textures (images) are used to create a photorealistic look. Aside from the color information, roughness, displacement (height differences), normal (surface direction), and ambient occulsion (small color brightness changes) are needed. These maps are also images, but instead of color, they represent their respective qualities.
This neural network generates the displacement, roughness, or ambient occulsion maps from the color map. The normal map can be generated from the displacement with another algorithm (not provided).
