Semantic StyleGAN: Latent Space Manipulation

Semantic manipulation of images through StyleGAN's latent space

Project Overview

This project explores semantic manipulation of images in StyleGAN2’s latent space, implementing several methods from the literature including InterfaceGAN, GANSpace, and Style Mixing. The goal is to give semantic meaning to the latent space of a pretrained StyleGAN2 model trained on the FFHQ dataset (faces).

Key Contributions

  • Implementation of InterfaceGAN for semantic attribute editing
  • GANSpace method for principal direction analysis
  • Style mixing for compositional image generation
  • Real image projection into latent space (W)
  • Integration of multiple state-of-the-art techniques

Technologies

  • Language: Python, CUDA, C++
  • Framework: PyTorch
  • Techniques: StyleGAN2, InterfaceGAN, GANSpace, SVM, PCA

Project Report

View or download the full project report:

Semantic StyleGAN Report

Repository

View on GitHub