Animated avatars are a long time part of our lives. But the realistic modeling of the entertainment fence remained an open challenge.
On the one hand, modern physical modeling techniques can generate realistic clothing geometry at interactive speed. On the other hand, the modeling of a photorealistic aspect generally requires a physical rendering, which is too expensive for interactive applications.
Colleagues from Meta Reality Labs have created a neural network that simulates “neural clothes” that work in addition to explicit physical geometry. The network is capable of generating a photorealistic appearance with dependent effects even of the invisible folds of clothes on the body.
Scientists have managed to achieve a photorealistic animated result even for loose clothes. But the key point is that the neural network allows you to “dress” the avatar in different clothes in real time.
Learn more details in the article https://arxiv.org/pdf/2206.15470.pdf
