This GitHub repository contains interesting experiments showcasing the capabilities of the pretrained StyleGAN2 network, which was trained on the custom dancer dataset. The images used here are synthetic human images obtained by training on StyleGAN and StyleGAN2 models. The primary dataset used in this project consists of 3,000 high-quality images of dancers in various poses.
The dataset used for training and experimentation comprises 3,000 high-quality images of dancers in various poses. These images were carefully curated to provide a diverse range of dance movements and body positions. The dataset serves as the foundation for creating synthetic human images with the StyleGAN and StyleGAN2 models.
One of the exciting experiments conducted using the pretrained StyleGAN2 network is the ability to morph or transition between different poses of dancers. This functionality allows for seamless transformations from one dancer pose to another, from a Yellow background to a Blue one, and so on.
Another fascinating experiment involves generating interpolation images using the StyleGAN2 model. Interpolation involves creating a sequence of images that gradually transition from one pose to another. This can be used to visualize smooth transformations between poses and explore the latent space of the model.
StyleGAN Official Implementation