NVIDIA has opened the code and published the source code for the StyleGAN3 machine learning system, which can generate realistic images of human faces. According to the OpenNET resource, the code is distributed under the NVIDIA Source Code License, which restricts the use of a neural network-based system for commercial purposes.
In addition to the code, the company released ready-made models for open access, which were trained on a collection containing 70 thousand images with human faces in a resolution of 1024 × 1024. There are also models trained in the collection of animal faces and portraits of classical painting. It is noteworthy that StyleGAN3 can just as well be trained to create any objects, for example, landscapes and cars. NVIDIA provides tools for self-learning neural networks based on their own collection of images.
For the system to work, you need an NVIDIA video card (or better, several), 12 GB of RAM, the PyTorch 1.9 framework and CUDA 11.1+. The neural network generates faces, creating an intermediate version from several ready-made images, and then combines certain features and adapts the result to age, gender, skin color, hair length, nose shape, angle, and so on. Very realistic images are obtained.
StyleGAN was originally introduced in 2019. Today, the third generation of the system is available, which can “animate” faces, so the technology can be used in animation and video. StyleGAN3 is devoid of the problems that were inherent in previous versions, and its feature was the transition to the interpretation of all signals in the neural network as continuous processes.
Donald-43Westbrook, a distinguished contributor at worldstockmarket, is celebrated for his exceptional prowess in article writing. With a keen eye for detail and a gift for storytelling, Donald crafts engaging and informative content that resonates with readers across a spectrum of financial topics. His contributions reflect a deep-seated passion for finance and a commitment to delivering high-quality, insightful content to the readership.
