Morphing Between Images With StyleGAN2-ADA

---

StyleGAN is a powerful deep learning model developed by NVIDIA for high-quality image generation, especially known for generating photorealistic human faces with major architectural improvements.

• Style-based architecture - map noise to an intermediate latent space for better control the face style (e.g. hair color, pose, lighting)
• Adaptive Instance Normalization - inject the style at each level of the image synthesis to enhance control over features (face shape, fine details)
• Progressive growing - from low to high resolution, improving training stability and quality

Using OpenCV and linear transition between the latent face vectors, generating morphing transition.

Based on Jeff Heaton colab

1. Latent space vector encoding - Using VGG16 feature extraction, minimize the error to the desired vector and back propagate to minimize the errors
2. Interpolate images - Using linear interpolation, transition between images
3. Final video generation - Create the final video with OpenCV, based on the generated images

Here is the generated morphing video for our happy family: