StyleGAN K-Girl Image Generation
Overview
This project involves training a StyleGAN model to generate high-quality images of K-girls (Korean-style female faces). The dataset is curated to capture the distinct aesthetic features common in K-girl images, and StyleGAN is used to create photorealistic outputs.
Dataset
- The dataset consists of high-resolution images of K-girls.
- Images are preprocessed and resized to fit StyleGAN's input requirements.
- Data augmentation techniques such as flipping and color adjustments are applied.
Model Training
- Framework: TensorFlow with NVIDIA StyleGAN implementation.
- Hardware: Trained using NVIDIA Tesla T4.
- Loss Function: Uses non-saturating GAN loss with R1 regularization.
- Training Duration: Depends on the dataset size and computing power.