Forest fog rendering using generative adversarial networks

Producing a physically realistic rendering of forests with fog involves the simulation of light diffusion in participating media. This volumetric phenomenon increases the realism of natural scenes in several fields, such as video games and flight simulators. However, this effect in a natural scene is characterized by complexity at all scales, representing a challenge for the computer graphics community. In this paper, we propose a novel approach based on a generative adversarial network to estimate fog in forest scenes. Our approach operates in two steps. The first step consists of training a generative adversarial network on a large dataset of synthetic images. Our network takes four images as input (normal map, depth map, albedo map and RGB map without fog), and it generates an estimation of the rendering forest with fog for output. A reference image conditions the input images to produce a better approximation. The second step consists of the production of realistic images with fog. Our technique can be generalized to high-frequency lighting effects (specularity and shadow), so it does not require any calculation in 3D space.